RFC 6787






Internet Engineering Task Force (IETF)                        D. Burnett
Request for Comments: 6787                                         Voxeo
Category: Standards Track                                  S. Shanmugham
ISSN: 2070-1721                                      Cisco Systems, Inc.
                                                           November 2012


           Media Resource Control Protocol Version 2 (MRCPv2)

Abstract



   The Media Resource Control Protocol Version 2 (MRCPv2) allows client
   hosts to control media service resources such as speech synthesizers,
   recognizers, verifiers, and identifiers residing in servers on the
   network.  MRCPv2 is not a "stand-alone" protocol -- it relies on
   other protocols, such as the Session Initiation Protocol (SIP), to
   coordinate MRCPv2 clients and servers and manage sessions between
   them, and the Session Description Protocol (SDP) to describe,
   discover, and exchange capabilities.  It also depends on SIP and SDP
   to establish the media sessions and associated parameters between the
   media source or sink and the media server.  Once this is done, the
   MRCPv2 exchange operates over the control session established above,
   allowing the client to control the media processing resources on the
   speech resource server.

Status of This Memo



   This is an Internet Standards Track document.

   This document is a product of the Internet Engineering Task Force
   (IETF).  It represents the consensus of the IETF community.  It has
   received public review and has been approved for publication by the
   Internet Engineering Steering Group (IESG).  Further information on
   Internet Standards is available in Section 2 of RFC 5741.

   Information about the current status of this document, any errata,
   and how to provide feedback on it may be obtained at
   http://www.rfc-editor.org/info/rfc6787.

Copyright Notice



   Copyright (c) 2012 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents
   (http://trustee.ietf.org/license-info) in effect on the date of
   publication of this document.  Please review these documents



Burnett & Shanmugham         Standards Track                    [Page 1]

RFC 6787                         MRCPv2                    November 2012


   carefully, as they describe your rights and restrictions with respect
   to this document.  Code Components extracted from this document must
   include Simplified BSD License text as described in Section 4.e of
   the Trust Legal Provisions and are provided without warranty as
   described in the Simplified BSD License.

   This document may contain material from IETF Documents or IETF
   Contributions published or made publicly available before November
   10, 2008.  The person(s) controlling the copyright in some of this
   material may not have granted the IETF Trust the right to allow
   modifications of such material outside the IETF Standards Process.
   Without obtaining an adequate license from the person(s) controlling
   the copyright in such materials, this document may not be modified
   outside the IETF Standards Process, and derivative works of it may
   not be created outside the IETF Standards Process, except to format
   it for publication as an RFC or to translate it into languages other
   than English.

Table of Contents



   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   8
   2.  Document Conventions  . . . . . . . . . . . . . . . . . . . .   9
     2.1.   Definitions  . . . . . . . . . . . . . . . . . . . . . .  10
     2.2.   State-Machine Diagrams . . . . . . . . . . . . . . . . .  10
     2.3.   URI Schemes  . . . . . . . . . . . . . . . . . . . . . .  11
   3.  Architecture  . . . . . . . . . . . . . . . . . . . . . . . .  11
     3.1.   MRCPv2 Media Resource Types  . . . . . . . . . . . . . .  12
     3.2.   Server and Resource Addressing . . . . . . . . . . . . .  14
   4.  MRCPv2 Basics . . . . . . . . . . . . . . . . . . . . . . . .  14
     4.1.   Connecting to the Server . . . . . . . . . . . . . . . .  14
     4.2.   Managing Resource Control Channels . . . . . . . . . . .  14
     4.3.   SIP Session Example  . . . . . . . . . . . . . . . . . .  17
     4.4.   Media Streams and RTP Ports  . . . . . . . . . . . . . .  22
     4.5.   MRCPv2 Message Transport . . . . . . . . . . . . . . . .  24
     4.6.   MRCPv2 Session Termination . . . . . . . . . . . . . . .  24
   5.  MRCPv2 Specification  . . . . . . . . . . . . . . . . . . . .  24
     5.1.   Common Protocol Elements . . . . . . . . . . . . . . . .  25
     5.2.   Request  . . . . . . . . . . . . . . . . . . . . . . . .  28
     5.3.   Response . . . . . . . . . . . . . . . . . . . . . . . .  29
     5.4.   Status Codes . . . . . . . . . . . . . . . . . . . . . .  30
     5.5.   Events . . . . . . . . . . . . . . . . . . . . . . . . .  31
   6.  MRCPv2 Generic Methods, Headers, and Result Structure . . . .  32
     6.1.   Generic Methods  . . . . . . . . . . . . . . . . . . . .  32
       6.1.1.   SET-PARAMS . . . . . . . . . . . . . . . . . . . . .  32
       6.1.2.   GET-PARAMS . . . . . . . . . . . . . . . . . . . . .  33
     6.2.   Generic Message Headers  . . . . . . . . . . . . . . . .  34
       6.2.1.   Channel-Identifier . . . . . . . . . . . . . . . . .  35
       6.2.2.   Accept . . . . . . . . . . . . . . . . . . . . . . .  36



Burnett & Shanmugham         Standards Track                    [Page 2]

RFC 6787                         MRCPv2                    November 2012


       6.2.3.   Active-Request-Id-List . . . . . . . . . . . . . . .  36
       6.2.4.   Proxy-Sync-Id  . . . . . . . . . . . . . . . . . . .  36
       6.2.5.   Accept-Charset . . . . . . . . . . . . . . . . . . .  37
       6.2.6.   Content-Type . . . . . . . . . . . . . . . . . . . .  37
       6.2.7.   Content-ID . . . . . . . . . . . . . . . . . . . . .  38
       6.2.8.   Content-Base . . . . . . . . . . . . . . . . . . . .  38
       6.2.9.   Content-Encoding . . . . . . . . . . . . . . . . . .  38
       6.2.10.  Content-Location . . . . . . . . . . . . . . . . . .  39
       6.2.11.  Content-Length . . . . . . . . . . . . . . . . . . .  39
       6.2.12.  Fetch Timeout  . . . . . . . . . . . . . . . . . . .  39
       6.2.13.  Cache-Control  . . . . . . . . . . . . . . . . . . .  40
       6.2.14.  Logging-Tag  . . . . . . . . . . . . . . . . . . . .  41
       6.2.15.  Set-Cookie . . . . . . . . . . . . . . . . . . . . .  42
       6.2.16.  Vendor-Specific Parameters . . . . . . . . . . . . .  44
     6.3.   Generic Result Structure . . . . . . . . . . . . . . . .  44
       6.3.1.   Natural Language Semantics Markup Language . . . . .  45
   7.  Resource Discovery  . . . . . . . . . . . . . . . . . . . . .  46
   8.  Speech Synthesizer Resource . . . . . . . . . . . . . . . . .  47
     8.1.   Synthesizer State Machine  . . . . . . . . . . . . . . .  48
     8.2.   Synthesizer Methods  . . . . . . . . . . . . . . . . . .  48
     8.3.   Synthesizer Events . . . . . . . . . . . . . . . . . . .  49
     8.4.   Synthesizer Header Fields  . . . . . . . . . . . . . . .  49
       8.4.1.   Jump-Size  . . . . . . . . . . . . . . . . . . . . .  49
       8.4.2.   Kill-On-Barge-In . . . . . . . . . . . . . . . . . .  50
       8.4.3.   Speaker-Profile  . . . . . . . . . . . . . . . . . .  51
       8.4.4.   Completion-Cause . . . . . . . . . . . . . . . . . .  51
       8.4.5.   Completion-Reason  . . . . . . . . . . . . . . . . .  52
       8.4.6.   Voice-Parameter  . . . . . . . . . . . . . . . . . .  52
       8.4.7.   Prosody-Parameters . . . . . . . . . . . . . . . . .  53
       8.4.8.   Speech-Marker  . . . . . . . . . . . . . . . . . . .  53
       8.4.9.   Speech-Language  . . . . . . . . . . . . . . . . . .  54
       8.4.10.  Fetch-Hint . . . . . . . . . . . . . . . . . . . . .  54
       8.4.11.  Audio-Fetch-Hint . . . . . . . . . . . . . . . . . .  55
       8.4.12.  Failed-URI . . . . . . . . . . . . . . . . . . . . .  55
       8.4.13.  Failed-URI-Cause . . . . . . . . . . . . . . . . . .  55
       8.4.14.  Speak-Restart  . . . . . . . . . . . . . . . . . . .  56
       8.4.15.  Speak-Length . . . . . . . . . . . . . . . . . . . .  56
       8.4.16.  Load-Lexicon . . . . . . . . . . . . . . . . . . . .  57
       8.4.17.  Lexicon-Search-Order . . . . . . . . . . . . . . . .  57
     8.5.   Synthesizer Message Body . . . . . . . . . . . . . . . .  57
       8.5.1.   Synthesizer Speech Data  . . . . . . . . . . . . . .  57
       8.5.2.   Lexicon Data . . . . . . . . . . . . . . . . . . . .  59
     8.6.   SPEAK Method . . . . . . . . . . . . . . . . . . . . . .  60
     8.7.   STOP . . . . . . . . . . . . . . . . . . . . . . . . . .  62
     8.8.   BARGE-IN-OCCURRED  . . . . . . . . . . . . . . . . . . .  63
     8.9.   PAUSE  . . . . . . . . . . . . . . . . . . . . . . . . .  65
     8.10.  RESUME . . . . . . . . . . . . . . . . . . . . . . . . .  66
     8.11.  CONTROL  . . . . . . . . . . . . . . . . . . . . . . . .  67



Burnett & Shanmugham         Standards Track                    [Page 3]

RFC 6787                         MRCPv2                    November 2012


     8.12.  SPEAK-COMPLETE . . . . . . . . . . . . . . . . . . . . .  69
     8.13.  SPEECH-MARKER  . . . . . . . . . . . . . . . . . . . . .  70
     8.14.  DEFINE-LEXICON . . . . . . . . . . . . . . . . . . . . .  71
   9.  Speech Recognizer Resource  . . . . . . . . . . . . . . . . .  72
     9.1.   Recognizer State Machine . . . . . . . . . . . . . . . .  74
     9.2.   Recognizer Methods . . . . . . . . . . . . . . . . . . .  74
     9.3.   Recognizer Events  . . . . . . . . . . . . . . . . . . .  75
     9.4.   Recognizer Header Fields . . . . . . . . . . . . . . . .  75
       9.4.1.   Confidence-Threshold . . . . . . . . . . . . . . . .  77
       9.4.2.   Sensitivity-Level  . . . . . . . . . . . . . . . . .  77
       9.4.3.   Speed-Vs-Accuracy  . . . . . . . . . . . . . . . . .  77
       9.4.4.   N-Best-List-Length . . . . . . . . . . . . . . . . .  78
       9.4.5.   Input-Type . . . . . . . . . . . . . . . . . . . . .  78
       9.4.6.   No-Input-Timeout . . . . . . . . . . . . . . . . . .  78
       9.4.7.   Recognition-Timeout  . . . . . . . . . . . . . . . .  79
       9.4.8.   Waveform-URI . . . . . . . . . . . . . . . . . . . .  79
       9.4.9.   Media-Type . . . . . . . . . . . . . . . . . . . . .  80
       9.4.10.  Input-Waveform-URI . . . . . . . . . . . . . . . . .  80
       9.4.11.  Completion-Cause . . . . . . . . . . . . . . . . . .  80
       9.4.12.  Completion-Reason  . . . . . . . . . . . . . . . . .  83
       9.4.13.  Recognizer-Context-Block . . . . . . . . . . . . . .  83
       9.4.14.  Start-Input-Timers . . . . . . . . . . . . . . . . .  83
       9.4.15.  Speech-Complete-Timeout  . . . . . . . . . . . . . .  84
       9.4.16.  Speech-Incomplete-Timeout  . . . . . . . . . . . . .  84
       9.4.17.  DTMF-Interdigit-Timeout  . . . . . . . . . . . . . .  85
       9.4.18.  DTMF-Term-Timeout  . . . . . . . . . . . . . . . . .  85
       9.4.19.  DTMF-Term-Char . . . . . . . . . . . . . . . . . . .  85
       9.4.20.  Failed-URI . . . . . . . . . . . . . . . . . . . . .  86
       9.4.21.  Failed-URI-Cause . . . . . . . . . . . . . . . . . .  86
       9.4.22.  Save-Waveform  . . . . . . . . . . . . . . . . . . .  86
       9.4.23.  New-Audio-Channel  . . . . . . . . . . . . . . . . .  86
       9.4.24.  Speech-Language  . . . . . . . . . . . . . . . . . .  87
       9.4.25.  Ver-Buffer-Utterance . . . . . . . . . . . . . . . .  87
       9.4.26.  Recognition-Mode . . . . . . . . . . . . . . . . . .  87
       9.4.27.  Cancel-If-Queue  . . . . . . . . . . . . . . . . . .  88
       9.4.28.  Hotword-Max-Duration . . . . . . . . . . . . . . . .  88
       9.4.29.  Hotword-Min-Duration . . . . . . . . . . . . . . . .  88
       9.4.30.  Interpret-Text . . . . . . . . . . . . . . . . . . .  89
       9.4.31.  DTMF-Buffer-Time . . . . . . . . . . . . . . . . . .  89
       9.4.32.  Clear-DTMF-Buffer  . . . . . . . . . . . . . . . . .  89
       9.4.33.  Early-No-Match . . . . . . . . . . . . . . . . . . .  90
       9.4.34.  Num-Min-Consistent-Pronunciations  . . . . . . . . .  90
       9.4.35.  Consistency-Threshold  . . . . . . . . . . . . . . .  90
       9.4.36.  Clash-Threshold  . . . . . . . . . . . . . . . . . .  90
       9.4.37.  Personal-Grammar-URI . . . . . . . . . . . . . . . .  91
       9.4.38.  Enroll-Utterance . . . . . . . . . . . . . . . . . .  91
       9.4.39.  Phrase-Id  . . . . . . . . . . . . . . . . . . . . .  91
       9.4.40.  Phrase-NL  . . . . . . . . . . . . . . . . . . . . .  92



Burnett & Shanmugham         Standards Track                    [Page 4]

RFC 6787                         MRCPv2                    November 2012


       9.4.41.  Weight . . . . . . . . . . . . . . . . . . . . . . .  92
       9.4.42.  Save-Best-Waveform . . . . . . . . . . . . . . . . .  92
       9.4.43.  New-Phrase-Id  . . . . . . . . . . . . . . . . . . .  93
       9.4.44.  Confusable-Phrases-URI . . . . . . . . . . . . . . .  93
       9.4.45.  Abort-Phrase-Enrollment  . . . . . . . . . . . . . .  93
     9.5.   Recognizer Message Body  . . . . . . . . . . . . . . . .  93
       9.5.1.   Recognizer Grammar Data  . . . . . . . . . . . . . .  93
       9.5.2.   Recognizer Result Data . . . . . . . . . . . . . . .  97
       9.5.3.   Enrollment Result Data . . . . . . . . . . . . . . .  98
       9.5.4.   Recognizer Context Block . . . . . . . . . . . . . .  98
     9.6.   Recognizer Results . . . . . . . . . . . . . . . . . . .  99
       9.6.1.   Markup Functions . . . . . . . . . . . . . . . . . .  99
       9.6.2.   Overview of Recognizer Result Elements and Their
                Relationships  . . . . . . . . . . . . . . . . . . . 100
       9.6.3.   Elements and Attributes  . . . . . . . . . . . . . . 101
     9.7.   Enrollment Results . . . . . . . . . . . . . . . . . . . 106
       9.7.1.   <num-clashes> Element  . . . . . . . . . . . . . . . 106
       9.7.2.   <num-good-repetitions> Element . . . . . . . . . . . 106
       9.7.3.   <num-repetitions-still-needed> Element . . . . . . . 107
       9.7.4.   <consistency-status> Element . . . . . . . . . . . . 107
       9.7.5.   <clash-phrase-ids> Element . . . . . . . . . . . . . 107
       9.7.6.   <transcriptions> Element . . . . . . . . . . . . . . 107
       9.7.7.   <confusable-phrases> Element . . . . . . . . . . . . 107
     9.8.   DEFINE-GRAMMAR . . . . . . . . . . . . . . . . . . . . . 107
     9.9.   RECOGNIZE  . . . . . . . . . . . . . . . . . . . . . . . 111
     9.10.  STOP . . . . . . . . . . . . . . . . . . . . . . . . . . 118
     9.11.  GET-RESULT . . . . . . . . . . . . . . . . . . . . . . . 119
     9.12.  START-OF-INPUT . . . . . . . . . . . . . . . . . . . . . 120
     9.13.  START-INPUT-TIMERS . . . . . . . . . . . . . . . . . . . 120
     9.14.  RECOGNITION-COMPLETE . . . . . . . . . . . . . . . . . . 120
     9.15.  START-PHRASE-ENROLLMENT  . . . . . . . . . . . . . . . . 123
     9.16.  ENROLLMENT-ROLLBACK  . . . . . . . . . . . . . . . . . . 124
     9.17.  END-PHRASE-ENROLLMENT  . . . . . . . . . . . . . . . . . 124
     9.18.  MODIFY-PHRASE  . . . . . . . . . . . . . . . . . . . . . 125
     9.19.  DELETE-PHRASE  . . . . . . . . . . . . . . . . . . . . . 125
     9.20.  INTERPRET  . . . . . . . . . . . . . . . . . . . . . . . 125
     9.21.  INTERPRETATION-COMPLETE  . . . . . . . . . . . . . . . . 127
     9.22.  DTMF Detection . . . . . . . . . . . . . . . . . . . . . 128
   10. Recorder Resource . . . . . . . . . . . . . . . . . . . . . . 129
     10.1.  Recorder State Machine . . . . . . . . . . . . . . . . . 129
     10.2.  Recorder Methods . . . . . . . . . . . . . . . . . . . . 130
     10.3.  Recorder Events  . . . . . . . . . . . . . . . . . . . . 130
     10.4.  Recorder Header Fields . . . . . . . . . . . . . . . . . 130
       10.4.1.  Sensitivity-Level  . . . . . . . . . . . . . . . . . 130
       10.4.2.  No-Input-Timeout . . . . . . . . . . . . . . . . . . 131
       10.4.3.  Completion-Cause . . . . . . . . . . . . . . . . . . 131
       10.4.4.  Completion-Reason  . . . . . . . . . . . . . . . . . 132
       10.4.5.  Failed-URI . . . . . . . . . . . . . . . . . . . . . 132



Burnett & Shanmugham         Standards Track                    [Page 5]

RFC 6787                         MRCPv2                    November 2012


       10.4.6.  Failed-URI-Cause . . . . . . . . . . . . . . . . . . 132
       10.4.7.  Record-URI . . . . . . . . . . . . . . . . . . . . . 132
       10.4.8.  Media-Type . . . . . . . . . . . . . . . . . . . . . 133
       10.4.9.  Max-Time . . . . . . . . . . . . . . . . . . . . . . 133
       10.4.10. Trim-Length  . . . . . . . . . . . . . . . . . . . . 134
       10.4.11. Final-Silence  . . . . . . . . . . . . . . . . . . . 134
       10.4.12. Capture-On-Speech  . . . . . . . . . . . . . . . . . 134
       10.4.13. Ver-Buffer-Utterance . . . . . . . . . . . . . . . . 134
       10.4.14. Start-Input-Timers . . . . . . . . . . . . . . . . . 135
       10.4.15. New-Audio-Channel  . . . . . . . . . . . . . . . . . 135
     10.5.  Recorder Message Body  . . . . . . . . . . . . . . . . . 135
     10.6.  RECORD . . . . . . . . . . . . . . . . . . . . . . . . . 135
     10.7.  STOP . . . . . . . . . . . . . . . . . . . . . . . . . . 136
     10.8.  RECORD-COMPLETE  . . . . . . . . . . . . . . . . . . . . 137
     10.9.  START-INPUT-TIMERS . . . . . . . . . . . . . . . . . . . 138
     10.10. START-OF-INPUT . . . . . . . . . . . . . . . . . . . . . 138
   11. Speaker Verification and Identification . . . . . . . . . . . 139
     11.1.  Speaker Verification State Machine . . . . . . . . . . . 140
     11.2.  Speaker Verification Methods . . . . . . . . . . . . . . 142
     11.3.  Verification Events  . . . . . . . . . . . . . . . . . . 144
     11.4.  Verification Header Fields . . . . . . . . . . . . . . . 144
       11.4.1.  Repository-URI . . . . . . . . . . . . . . . . . . . 144
       11.4.2.  Voiceprint-Identifier  . . . . . . . . . . . . . . . 145
       11.4.3.  Verification-Mode  . . . . . . . . . . . . . . . . . 145
       11.4.4.  Adapt-Model  . . . . . . . . . . . . . . . . . . . . 146
       11.4.5.  Abort-Model  . . . . . . . . . . . . . . . . . . . . 146
       11.4.6.  Min-Verification-Score . . . . . . . . . . . . . . . 147
       11.4.7.  Num-Min-Verification-Phrases . . . . . . . . . . . . 147
       11.4.8.  Num-Max-Verification-Phrases . . . . . . . . . . . . 147
       11.4.9.  No-Input-Timeout . . . . . . . . . . . . . . . . . . 148
       11.4.10. Save-Waveform  . . . . . . . . . . . . . . . . . . . 148
       11.4.11. Media-Type . . . . . . . . . . . . . . . . . . . . . 148
       11.4.12. Waveform-URI . . . . . . . . . . . . . . . . . . . . 148
       11.4.13. Voiceprint-Exists  . . . . . . . . . . . . . . . . . 149
       11.4.14. Ver-Buffer-Utterance . . . . . . . . . . . . . . . . 149
       11.4.15. Input-Waveform-URI . . . . . . . . . . . . . . . . . 149
       11.4.16. Completion-Cause . . . . . . . . . . . . . . . . . . 150
       11.4.17. Completion-Reason  . . . . . . . . . . . . . . . . . 151
       11.4.18. Speech-Complete-Timeout  . . . . . . . . . . . . . . 151
       11.4.19. New-Audio-Channel  . . . . . . . . . . . . . . . . . 152
       11.4.20. Abort-Verification . . . . . . . . . . . . . . . . . 152
       11.4.21. Start-Input-Timers . . . . . . . . . . . . . . . . . 152
     11.5.  Verification Message Body  . . . . . . . . . . . . . . . 152
       11.5.1.  Verification Result Data . . . . . . . . . . . . . . 152
       11.5.2.  Verification Result Elements . . . . . . . . . . . . 153
     11.6.  START-SESSION  . . . . . . . . . . . . . . . . . . . . . 157
     11.7.  END-SESSION  . . . . . . . . . . . . . . . . . . . . . . 158
     11.8.  QUERY-VOICEPRINT . . . . . . . . . . . . . . . . . . . . 159



Burnett & Shanmugham         Standards Track                    [Page 6]

RFC 6787                         MRCPv2                    November 2012


     11.9.  DELETE-VOICEPRINT  . . . . . . . . . . . . . . . . . . . 160
     11.10. VERIFY . . . . . . . . . . . . . . . . . . . . . . . . . 160
     11.11. VERIFY-FROM-BUFFER . . . . . . . . . . . . . . . . . . . 160
     11.12. VERIFY-ROLLBACK  . . . . . . . . . . . . . . . . . . . . 164
     11.13. STOP . . . . . . . . . . . . . . . . . . . . . . . . . . 164
     11.14. START-INPUT-TIMERS . . . . . . . . . . . . . . . . . . . 165
     11.15. VERIFICATION-COMPLETE  . . . . . . . . . . . . . . . . . 165
     11.16. START-OF-INPUT . . . . . . . . . . . . . . . . . . . . . 166
     11.17. CLEAR-BUFFER . . . . . . . . . . . . . . . . . . . . . . 166
     11.18. GET-INTERMEDIATE-RESULT  . . . . . . . . . . . . . . . . 167
   12. Security Considerations . . . . . . . . . . . . . . . . . . . 168
     12.1.  Rendezvous and Session Establishment . . . . . . . . . . 168
     12.2.  Control Channel Protection . . . . . . . . . . . . . . . 168
     12.3.  Media Session Protection . . . . . . . . . . . . . . . . 169
     12.4.  Indirect Content Access  . . . . . . . . . . . . . . . . 169
     12.5.  Protection of Stored Media . . . . . . . . . . . . . . . 170
     12.6.  DTMF and Recognition Buffers . . . . . . . . . . . . . . 171
     12.7.  Client-Set Server Parameters . . . . . . . . . . . . . . 171
     12.8.  DELETE-VOICEPRINT and Authorization  . . . . . . . . . . 171
   13. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 171
     13.1.  New Registries . . . . . . . . . . . . . . . . . . . . . 171
       13.1.1.  MRCPv2 Resource Types  . . . . . . . . . . . . . . . 171
       13.1.2.  MRCPv2 Methods and Events  . . . . . . . . . . . . . 172
       13.1.3.  MRCPv2 Header Fields . . . . . . . . . . . . . . . . 173
       13.1.4.  MRCPv2 Status Codes  . . . . . . . . . . . . . . . . 176
       13.1.5.  Grammar Reference List Parameters  . . . . . . . . . 176
       13.1.6.  MRCPv2 Vendor-Specific Parameters  . . . . . . . . . 176
     13.2.  NLSML-Related Registrations  . . . . . . . . . . . . . . 177
       13.2.1.  'application/nlsml+xml' Media Type Registration  . . 177
     13.3.  NLSML XML Schema Registration  . . . . . . . . . . . . . 178
     13.4.  MRCPv2 XML Namespace Registration  . . . . . . . . . . . 178
     13.5.  Text Media Type Registrations  . . . . . . . . . . . . . 178
       13.5.1.  text/grammar-ref-list  . . . . . . . . . . . . . . . 178
     13.6.  'session' URI Scheme Registration  . . . . . . . . . . . 180
     13.7.  SDP Parameter Registrations  . . . . . . . . . . . . . . 181
       13.7.1.  Sub-Registry "proto" . . . . . . . . . . . . . . . . 181
       13.7.2.  Sub-Registry "att-field (media-level)" . . . . . . . 182
   14. Examples  . . . . . . . . . . . . . . . . . . . . . . . . . . 183
     14.1.  Message Flow . . . . . . . . . . . . . . . . . . . . . . 183
     14.2.  Recognition Result Examples  . . . . . . . . . . . . . . 192
       14.2.1.  Simple ASR Ambiguity . . . . . . . . . . . . . . . . 192
       14.2.2.  Mixed Initiative . . . . . . . . . . . . . . . . . . 192
       14.2.3.  DTMF Input . . . . . . . . . . . . . . . . . . . . . 193
       14.2.4.  Interpreting Meta-Dialog and Meta-Task Utterances  . 194
       14.2.5.  Anaphora and Deixis  . . . . . . . . . . . . . . . . 195
       14.2.6.  Distinguishing Individual Items from Sets with
                One Member . . . . . . . . . . . . . . . . . . . . . 195
       14.2.7.  Extensibility  . . . . . . . . . . . . . . . . . . . 196



Burnett & Shanmugham         Standards Track                    [Page 7]

RFC 6787                         MRCPv2                    November 2012


   15. ABNF Normative Definition . . . . . . . . . . . . . . . . . . 196
   16. XML Schemas . . . . . . . . . . . . . . . . . . . . . . . . . 211
     16.1.  NLSML Schema Definition  . . . . . . . . . . . . . . . . 211
     16.2.  Enrollment Results Schema Definition . . . . . . . . . . 213
     16.3.  Verification Results Schema Definition . . . . . . . . . 214
   17. References  . . . . . . . . . . . . . . . . . . . . . . . . . 218
     17.1.  Normative References . . . . . . . . . . . . . . . . . . 218
     17.2.  Informative References . . . . . . . . . . . . . . . . . 220
   Appendix A.  Contributors . . . . . . . . . . . . . . . . . . . . 223
   Appendix B.  Acknowledgements . . . . . . . . . . . . . . . . . . 223

1.  Introduction



   MRCPv2 is designed to allow a client device to control media
   processing resources on the network.  Some of these media processing
   resources include speech recognition engines, speech synthesis
   engines, speaker verification, and speaker identification engines.
   MRCPv2 enables the implementation of distributed Interactive Voice
   Response platforms using VoiceXML [W3C.REC-voicexml20-20040316]
   browsers or other client applications while maintaining separate
   back-end speech processing capabilities on specialized speech
   processing servers.  MRCPv2 is based on the earlier Media Resource
   Control Protocol (MRCP) [RFC4463] developed jointly by Cisco Systems,
   Inc., Nuance Communications, and Speechworks, Inc.  Although some of
   the method names are similar, the way in which these methods are
   communicated is different.  There are also more resources and more
   methods for each resource.  The first version of MRCP was essentially
   taken only as input to the development of this protocol.  There is no
   expectation that an MRCPv2 client will work with an MRCPv1 server or
   vice versa.  There is no migration plan or gateway definition between
   the two protocols.

   The protocol requirements of Speech Services Control (SPEECHSC)
   [RFC4313] include that the solution be capable of reaching a media
   processing server, setting up communication channels to the media
   resources, and sending and receiving control messages and media
   streams to/from the server.  The Session Initiation Protocol (SIP)
   [RFC3261] meets these requirements.

   The proprietary version of MRCP ran over the Real Time Streaming
   Protocol (RTSP) [RFC2326].  At the time work on MRCPv2 was begun, the
   consensus was that this use of RTSP would break the RTSP protocol or
   cause backward-compatibility problems, something forbidden by Section
   3.2 of [RFC4313].  This is the reason why MRCPv2 does not run over
   RTSP.






Burnett & Shanmugham         Standards Track                    [Page 8]

RFC 6787                         MRCPv2                    November 2012


   MRCPv2 leverages these capabilities by building upon SIP and the
   Session Description Protocol (SDP) [RFC4566].  MRCPv2 uses SIP to set
   up and tear down media and control sessions with the server.  In
   addition, the client can use a SIP re-INVITE method (an INVITE dialog
   sent within an existing SIP session) to change the characteristics of
   these media and control session while maintaining the SIP dialog
   between the client and server.  SDP is used to describe the
   parameters of the media sessions associated with that dialog.  It is
   mandatory to support SIP as the session establishment protocol to
   ensure interoperability.  Other protocols can be used for session
   establishment by prior agreement.  This document only describes the
   use of SIP and SDP.

   MRCPv2 uses SIP and SDP to create the speech client/server dialog and
   set up the media channels to the server.  It also uses SIP and SDP to
   establish MRCPv2 control sessions between the client and the server
   for each media processing resource required for that dialog.  The
   MRCPv2 protocol exchange between the client and the media resource is
   carried on that control session.  MRCPv2 exchanges do not change the
   state of the SIP dialog, the media sessions, or other parameters of
   the dialog initiated via SIP.  It controls and affects the state of
   the media processing resource associated with the MRCPv2 session(s).

   MRCPv2 defines the messages to control the different media processing
   resources and the state machines required to guide their operation.
   It also describes how these messages are carried over a transport-
   layer protocol such as the Transmission Control Protocol (TCP)
   [RFC0793] or the Transport Layer Security (TLS) Protocol [RFC5246].
   (Note: the Stream Control Transmission Protocol (SCTP) [RFC4960] is a
   viable transport for MRCPv2 as well, but the mapping onto SCTP is not
   described in this specification.)

2.  Document Conventions



   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
   document are to be interpreted as described in RFC 2119 [RFC2119].

   Since many of the definitions and syntax are identical to those for
   the Hypertext Transfer Protocol -- HTTP/1.1 [RFC2616], this
   specification refers to the section where they are defined rather
   than copying it.  For brevity, [HX.Y] is to be taken to refer to
   Section X.Y of RFC 2616.

   All the mechanisms specified in this document are described in both
   prose and an augmented Backus-Naur form (ABNF [RFC5234]).





Burnett & Shanmugham         Standards Track                    [Page 9]

RFC 6787                         MRCPv2                    November 2012


   The complete message format in ABNF form is provided in Section 15
   and is the normative format definition.  Note that productions may be
   duplicated within the main body of the document for reading
   convenience.  If a production in the body of the text conflicts with
   one in the normative definition, the latter rules.

2.1.  Definitions



   Media Resource
                  An entity on the speech processing server that can be
                  controlled through MRCPv2.

   MRCP Server
                  Aggregate of one or more "Media Resource" entities on
                  a server, exposed through MRCPv2.  Often, 'server' in
                  this document refers to an MRCP server.

   MRCP Client
                  An entity controlling one or more Media Resources
                  through MRCPv2 ("Client" for short).

   DTMF
                  Dual-Tone Multi-Frequency; a method of transmitting
                  key presses in-band, either as actual tones (Q.23
                  [Q.23]) or as named tone events (RFC 4733 [RFC4733]).

   Endpointing
                  The process of automatically detecting the beginning
                  and end of speech in an audio stream.  This is
                  critical both for speech recognition and for automated
                  recording as one would find in voice mail systems.

   Hotword Mode
                  A mode of speech recognition where a stream of
                  utterances is evaluated for match against a small set
                  of command words.  This is generally employed either
                  to trigger some action or to control the subsequent
                  grammar to be used for further recognition.

2.2.  State-Machine Diagrams



   The state-machine diagrams in this document do not show every
   possible method call.  Rather, they reflect the state of the resource
   based on the methods that have moved to IN-PROGRESS or COMPLETE
   states (see Section 5.3).  Note that since PENDING requests
   essentially have not affected the resource yet and are in the queue
   to be processed, they are not reflected in the state-machine
   diagrams.



Burnett & Shanmugham         Standards Track                   [Page 10]

RFC 6787                         MRCPv2                    November 2012


2.3.  URI Schemes



   This document defines many protocol headers that contain URIs
   (Uniform Resource Identifiers [RFC3986]) or lists of URIs for
   referencing media.  The entire document, including the Security
   Considerations section (Section 12), assumes that HTTP or HTTP over
   TLS (HTTPS) [RFC2818] will be used as the URI addressing scheme
   unless otherwise stated.  However, implementations MAY support other
   schemes (such as 'file'), provided they have addressed any security
   considerations described in this document and any others particular
   to the specific scheme.  For example, implementations where the
   client and server both reside on the same physical hardware and the
   file system is secured by traditional user-level file access controls
   could be reasonable candidates for supporting the 'file' scheme.

3.  Architecture



   A system using MRCPv2 consists of a client that requires the
   generation and/or consumption of media streams and a media resource
   server that has the resources or "engines" to process these streams
   as input or generate these streams as output.  The client uses SIP
   and SDP to establish an MRCPv2 control channel with the server to use
   its media processing resources.  MRCPv2 servers are addressed using
   SIP URIs.

   SIP uses SDP with the offer/answer model described in RFC 3264
   [RFC3264] to set up the MRCPv2 control channels and describe their
   characteristics.  A separate MRCPv2 session is needed to control each
   of the media processing resources associated with the SIP dialog
   between the client and server.  Within a SIP dialog, the individual
   resource control channels for the different resources are added or
   removed through SDP offer/answer carried in a SIP re-INVITE
   transaction.

   The server, through the SDP exchange, provides the client with a
   difficult-to-guess, unambiguous channel identifier and a TCP port
   number (see Section 4.2).  The client MAY then open a new TCP
   connection with the server on this port number.  Multiple MRCPv2
   channels can share a TCP connection between the client and the
   server.  All MRCPv2 messages exchanged between the client and the
   server carry the specified channel identifier that the server MUST
   ensure is unambiguous among all MRCPv2 control channels that are
   active on that server.  The client uses this channel identifier to
   indicate the media processing resource associated with that channel.
   For information on message framing, see Section 5.

   SIP also establishes the media sessions between the client (or other
   source/sink of media) and the MRCPv2 server using SDP "m=" lines.



Burnett & Shanmugham         Standards Track                   [Page 11]

RFC 6787                         MRCPv2                    November 2012


   One or more media processing resources may share a media session
   under a SIP session, or each media processing resource may have its
   own media session.

   The following diagram shows the general architecture of a system that
   uses MRCPv2.  To simplify the diagram, only a few resources are
   shown.

     MRCPv2 client                   MRCPv2 Media Resource Server
|--------------------|            |------------------------------------|
||------------------||            ||----------------------------------||
|| Application Layer||            ||Synthesis|Recognition|Verification||
||------------------||            || Engine  |  Engine   |   Engine   ||
||Media Resource API||            ||    ||   |    ||     |    ||      ||
||------------------||            ||Synthesis|Recognizer |  Verifier  ||
|| SIP  |  MRCPv2   ||            ||Resource | Resource  |  Resource  ||
||Stack |           ||            ||     Media Resource Management    ||
||      |           ||            ||----------------------------------||
||------------------||            ||   SIP  |        MRCPv2           ||
||   TCP/IP Stack   ||---MRCPv2---||  Stack |                         ||
||                  ||            ||----------------------------------||
||------------------||----SIP-----||           TCP/IP Stack           ||
|--------------------|            ||                                  ||
         |                        ||----------------------------------||
        SIP                       |------------------------------------|
         |                          /
|-------------------|             RTP
|                   |             /
| Media Source/Sink |------------/
|                   |
|-------------------|

                      Figure 1: Architectural Diagram

3.1.  MRCPv2 Media Resource Types



   An MRCPv2 server may offer one or more of the following media
   processing resources to its clients.

   Basic Synthesizer
                  A speech synthesizer resource that has very limited
                  capabilities and can generate its media stream
                  exclusively from concatenated audio clips.  The speech
                  data is described using a limited subset of the Speech
                  Synthesis Markup Language (SSML)
                  [W3C.REC-speech-synthesis-20040907] elements.  A basic
                  synthesizer MUST support the SSML tags <speak>,
                  <audio>, <say-as>, and <mark>.



Burnett & Shanmugham         Standards Track                   [Page 12]

RFC 6787                         MRCPv2                    November 2012


   Speech Synthesizer
                  A full-capability speech synthesis resource that can
                  render speech from text.  Such a synthesizer MUST have
                  full SSML [W3C.REC-speech-synthesis-20040907] support.

   Recorder
                  A resource capable of recording audio and providing a
                  URI pointer to the recording.  A recorder MUST provide
                  endpointing capabilities for suppressing silence at
                  the beginning and end of a recording, and MAY also
                  suppress silence in the middle of a recording.  If
                  such suppression is done, the recorder MUST maintain
                  timing metadata to indicate the actual timestamps of
                  the recorded media.

   DTMF Recognizer
                  A recognizer resource capable of extracting and
                  interpreting Dual-Tone Multi-Frequency (DTMF) [Q.23]
                  digits in a media stream and matching them against a
                  supplied digit grammar.  It could also do a semantic
                  interpretation based on semantic tags in the grammar.

   Speech Recognizer
                  A full speech recognition resource that is capable of
                  receiving a media stream containing audio and
                  interpreting it to recognition results.  It also has a
                  natural language semantic interpreter to post-process
                  the recognized data according to the semantic data in
                  the grammar and provide semantic results along with
                  the recognized input.  The recognizer MAY also support
                  enrolled grammars, where the client can enroll and
                  create new personal grammars for use in future
                  recognition operations.

   Speaker Verifier
                  A resource capable of verifying the authenticity of a
                  claimed identity by matching a media stream containing
                  spoken input to a pre-existing voiceprint.  This may
                  also involve matching the caller's voice against more
                  than one voiceprint, also called multi-verification or
                  speaker identification.










Burnett & Shanmugham         Standards Track                   [Page 13]

RFC 6787                         MRCPv2                    November 2012


3.2.  Server and Resource Addressing



   The MRCPv2 server is a generic SIP server, and is thus addressed by a
   SIP URI (RFC 3261 [RFC3261]).

   For example:

        sip:mrcpv2@example.net   or
        sips:mrcpv2@example.net

4.  MRCPv2 Basics



   MRCPv2 requires a connection-oriented transport-layer protocol such
   as TCP to guarantee reliable sequencing and delivery of MRCPv2
   control messages between the client and the server.  In order to meet
   the requirements for security enumerated in SPEECHSC requirements
   [RFC4313], clients and servers MUST implement TLS as well.  One or
   more connections between the client and the server can be shared
   among different MRCPv2 channels to the server.  The individual
   messages carry the channel identifier to differentiate messages on
   different channels.  MRCPv2 encoding is text based with mechanisms to
   carry embedded binary data.  This allows arbitrary data like
   recognition grammars, recognition results, synthesizer speech markup,
   etc., to be carried in MRCPv2 messages.  For information on message
   framing, see Section 5.

4.1.  Connecting to the Server



   MRCPv2 employs SIP, in conjunction with SDP, as the session
   establishment and management protocol.  The client reaches an MRCPv2
   server using conventional INVITE and other SIP requests for
   establishing, maintaining, and terminating SIP dialogs.  The SDP
   offer/answer exchange model over SIP is used to establish a resource
   control channel for each resource.  The SDP offer/answer exchange is
   also used to establish media sessions between the server and the
   source or sink of audio.

4.2.  Managing Resource Control Channels



   The client needs a separate MRCPv2 resource control channel to
   control each media processing resource under the SIP dialog.  A
   unique channel identifier string identifies these resource control
   channels.  The channel identifier is a difficult-to-guess,
   unambiguous string followed by an "@", then by a string token
   specifying the type of resource.  The server generates the channel
   identifier and MUST make sure it does not clash with the identifier
   of any other MRCP channel currently allocated by that server.  MRCPv2
   defines the following IANA-registered types of media processing



Burnett & Shanmugham         Standards Track                   [Page 14]

RFC 6787                         MRCPv2                    November 2012


   resources.  Additional resource types and their associated methods/
   events and state machines may be added as described below in
   Section 13.

          +---------------+----------------------+--------------+
          | Resource Type | Resource Description | Described in |
          +---------------+----------------------+--------------+
          | speechrecog   | Speech Recognizer    | Section 9    |
          | dtmfrecog     | DTMF Recognizer      | Section 9    |
          | speechsynth   | Speech Synthesizer   | Section 8    |
          | basicsynth    | Basic Synthesizer    | Section 8    |
          | speakverify   | Speaker Verification | Section 11   |
          | recorder      | Speech Recorder      | Section 10   |
          +---------------+----------------------+--------------+

                          Table 1: Resource Types

   The SIP INVITE or re-INVITE transaction and the SDP offer/answer
   exchange it carries contain "m=" lines describing the resource
   control channel to be allocated.  There MUST be one SDP "m=" line for
   each MRCPv2 resource to be used in the session.  This "m=" line MUST
   have a media type field of "application" and a transport type field
   of either "TCP/MRCPv2" or "TCP/TLS/MRCPv2".  The port number field of
   the "m=" line MUST contain the "discard" port of the transport
   protocol (port 9 for TCP) in the SDP offer from the client and MUST
   contain the TCP listen port on the server in the SDP answer.  The
   client may then either set up a TCP or TLS connection to that server
   port or share an already established connection to that port.  Since
   MRCPv2 allows multiple sessions to share the same TCP connection,
   multiple "m=" lines in a single SDP document MAY share the same port
   field value; MRCPv2 servers MUST NOT assume any relationship between
   resources using the same port other than the sharing of the
   communication channel.

   MRCPv2 resources do not use the port or format field of the "m=" line
   to distinguish themselves from other resources using the same
   channel.  The client MUST specify the resource type identifier in the
   resource attribute associated with the control "m=" line of the SDP
   offer.  The server MUST respond with the full Channel-Identifier
   (which includes the resource type identifier and a difficult-to-
   guess, unambiguous string) in the "channel" attribute associated with
   the control "m=" line of the SDP answer.  To remain backwards
   compatible with conventional SDP usage, the format field of the "m="
   line MUST have the arbitrarily selected value of "1".

   When the client wants to add a media processing resource to the
   session, it issues a new SDP offer, according to the procedures of
   RFC 3264 [RFC3264], in a SIP re-INVITE request.  The SDP offer/answer



Burnett & Shanmugham         Standards Track                   [Page 15]

RFC 6787                         MRCPv2                    November 2012


   exchange carried by this SIP transaction contains one or more
   additional control "m=" lines for the new resources to be allocated
   to the session.  The server, on seeing the new "m=" line, allocates
   the resources (if they are available) and responds with a
   corresponding control "m=" line in the SDP answer carried in the SIP
   response.  If the new resources are not available, the re-INVITE
   receives an error message, and existing media processing going on
   before the re-INVITE will continue as it was before.  It is not
   possible to allocate more than one resource of each type.  If a
   client requests more than one resource of any type, the server MUST
   behave as if the resources of that type (beyond the first one) are
   not available.

   MRCPv2 clients and servers using TCP as a transport protocol MUST use
   the procedures specified in RFC 4145 [RFC4145] for setting up the TCP
   connection, with the considerations described hereby.  Similarly,
   MRCPv2 clients and servers using TCP/TLS as a transport protocol MUST
   use the procedures specified in RFC 4572 [RFC4572] for setting up the
   TLS connection, with the considerations described hereby.  The
   a=setup attribute, as described in RFC 4145 [RFC4145], MUST be
   "active" for the offer from the client and MUST be "passive" for the
   answer from the MRCPv2 server.  The a=connection attribute MUST have
   a value of "new" on the very first control "m=" line offer from the
   client to an MRCPv2 server.  Subsequent control "m=" line offers from
   the client to the MRCP server MAY contain "new" or "existing",
   depending on whether the client wants to set up a new connection or
   share an existing connection, respectively.  If the client specifies
   a value of "new", the server MUST respond with a value of "new".  If
   the client specifies a value of "existing", the server MUST respond.
   The legal values in the response are "existing" if the server prefers
   to share an existing connection or "new" if not.  In the latter case,
   the client MUST initiate a new transport connection.

   When the client wants to deallocate the resource from this session,
   it issues a new SDP offer, according to RFC 3264 [RFC3264], where the
   control "m=" line port MUST be set to 0.  This SDP offer is sent in a
   SIP re-INVITE request.  This deallocates the associated MRCPv2
   identifier and resource.  The server MUST NOT close the TCP or TLS
   connection if it is currently being shared among multiple MRCP
   channels.  When all MRCP channels that may be sharing the connection
   are released and/or the associated SIP dialog is terminated, the
   client or server terminates the connection.

   When the client wants to tear down the whole session and all its
   resources, it MUST issue a SIP BYE request to close the SIP session.
   This will deallocate all the control channels and resources allocated
   under the session.




Burnett & Shanmugham         Standards Track                   [Page 16]

RFC 6787                         MRCPv2                    November 2012


   All servers MUST support TLS.  Servers MAY use TCP without TLS in
   controlled environments (e.g., not in the public Internet) where both
   nodes are inside a protected perimeter, for example, preventing
   access to the MRCP server from remote nodes outside the controlled
   perimeter.  It is up to the client, through the SDP offer, to choose
   which transport it wants to use for an MRCPv2 session.  Aside from
   the exceptions given above, when using TCP, the "m=" lines MUST
   conform to RFC4145 [RFC4145], which describes the usage of SDP for
   connection-oriented transport.  When using TLS, the SDP "m=" line for
   the control stream MUST conform to Connection-Oriented Media
   (COMEDIA) over TLS [RFC4572], which specifies the usage of SDP for
   establishing a secure connection-oriented transport over TLS.

4.3.  SIP Session Example



   This first example shows the power of using SIP to route to the
   appropriate resource.  In the example, note the use of a request to a
   domain's speech server service in the INVITE to
   mresources@example.com.  The SIP routing machinery in the domain
   locates the actual server, mresources@server.example.com, which gets
   returned in the 200 OK.  Note that "cmid" is defined in Section 4.4.

   This example exchange adds a resource control channel for a
   synthesizer.  Since a synthesizer also generates an audio stream,
   this interaction also creates a receive-only Real-Time Protocol (RTP)
   [RFC3550] media session for the server to send audio to.  The SIP
   dialog with the media source/sink is independent of MRCP and is not
   shown.

   C->S:  INVITE sip:mresources@example.com SIP/2.0
          Via:SIP/2.0/TCP client.atlanta.example.com:5060;
           branch=z9hG4bK74bf1
          Max-Forwards:6
          To:MediaServer <sip:mresources@example.com>
          From:sarvi <sip:sarvi@example.com>;tag=1928301774
          Call-ID:a84b4c76e66710
          CSeq:314161 INVITE
          Contact:<sip:sarvi@client.example.com>
          Content-Type:application/sdp
          Content-Length:...

          v=0
          o=sarvi 2890844526 2890844526 IN IP4 192.0.2.12
          s=-
          c=IN IP4 192.0.2.12
          t=0 0
          m=application 9 TCP/MRCPv2 1
          a=setup:active



Burnett & Shanmugham         Standards Track                   [Page 17]

RFC 6787                         MRCPv2                    November 2012


          a=connection:new
          a=resource:speechsynth
          a=cmid:1
          m=audio 49170 RTP/AVP 0
          a=rtpmap:0 pcmu/8000
          a=recvonly
          a=mid:1


   S->C:  SIP/2.0 200 OK
          Via:SIP/2.0/TCP client.atlanta.example.com:5060;
           branch=z9hG4bK74bf1;received=192.0.32.10
          To:MediaServer <sip:mresources@example.com>;tag=62784
          From:sarvi <sip:sarvi@example.com>;tag=1928301774
          Call-ID:a84b4c76e66710
          CSeq:314161 INVITE
          Contact:<sip:mresources@server.example.com>
          Content-Type:application/sdp
          Content-Length:...

          v=0
          o=- 2890842808 2890842808 IN IP4 192.0.2.11
          s=-
          c=IN IP4 192.0.2.11
          t=0 0
          m=application 32416 TCP/MRCPv2 1
          a=setup:passive
          a=connection:new
          a=channel:32AECB234338@speechsynth
          a=cmid:1
          m=audio 48260 RTP/AVP 0
          a=rtpmap:0 pcmu/8000
          a=sendonly
          a=mid:1


   C->S:  ACK sip:mresources@server.example.com SIP/2.0
          Via:SIP/2.0/TCP client.atlanta.example.com:5060;
           branch=z9hG4bK74bf2
          Max-Forwards:6
          To:MediaServer <sip:mresources@example.com>;tag=62784
          From:Sarvi <sip:sarvi@example.com>;tag=1928301774
          Call-ID:a84b4c76e66710
          CSeq:314161 ACK
          Content-Length:0

                 Example: Add Synthesizer Control Channel




Burnett & Shanmugham         Standards Track                   [Page 18]

RFC 6787                         MRCPv2                    November 2012


   This example exchange continues from the previous figure and
   allocates an additional resource control channel for a recognizer.
   Since a recognizer would need to receive an audio stream for
   recognition, this interaction also updates the audio stream to
   sendrecv, making it a two-way RTP media session.

   C->S:  INVITE sip:mresources@server.example.com SIP/2.0
          Via:SIP/2.0/TCP client.atlanta.example.com:5060;
           branch=z9hG4bK74bf3
          Max-Forwards:6
          To:MediaServer <sip:mresources@example.com>;tag=62784
          From:sarvi <sip:sarvi@example.com>;tag=1928301774
          Call-ID:a84b4c76e66710
          CSeq:314162 INVITE
          Contact:<sip:sarvi@client.example.com>
          Content-Type:application/sdp
          Content-Length:...

          v=0
          o=sarvi 2890844526 2890844527 IN IP4 192.0.2.12
          s=-
          c=IN IP4 192.0.2.12
          t=0 0
          m=application 9 TCP/MRCPv2 1
          a=setup:active
          a=connection:existing
          a=resource:speechsynth
          a=cmid:1
          m=audio 49170 RTP/AVP 0 96
          a=rtpmap:0 pcmu/8000
          a=rtpmap:96 telephone-event/8000
          a=fmtp:96 0-15
          a=sendrecv
          a=mid:1
          m=application 9 TCP/MRCPv2 1
          a=setup:active
          a=connection:existing
          a=resource:speechrecog
          a=cmid:1


   S->C:  SIP/2.0 200 OK
          Via:SIP/2.0/TCP client.atlanta.example.com:5060;
           branch=z9hG4bK74bf3;received=192.0.32.10
          To:MediaServer <sip:mresources@example.com>;tag=62784
          From:sarvi <sip:sarvi@example.com>;tag=1928301774
          Call-ID:a84b4c76e66710
          CSeq:314162 INVITE



Burnett & Shanmugham         Standards Track                   [Page 19]

RFC 6787                         MRCPv2                    November 2012


          Contact:<sip:mresources@server.example.com>
          Content-Type:application/sdp
          Content-Length:...

          v=0
          o=- 2890842808 2890842809 IN IP4 192.0.2.11
          s=-
          c=IN IP4 192.0.2.11
          t=0 0
          m=application 32416 TCP/MRCPv2 1
          a=setup:passive
          a=connection:existing
          a=channel:32AECB234338@speechsynth
          a=cmid:1
          m=audio 48260 RTP/AVP 0 96
          a=rtpmap:0 pcmu/8000
          a=rtpmap:96 telephone-event/8000
          a=fmtp:96 0-15
          a=sendrecv
          a=mid:1
          m=application 32416 TCP/MRCPv2 1
          a=setup:passive
          a=connection:existing
          a=channel:32AECB234338@speechrecog
          a=cmid:1


   C->S:  ACK sip:mresources@server.example.com SIP/2.0
          Via:SIP/2.0/TCP client.atlanta.example.com:5060;
           branch=z9hG4bK74bf4
          Max-Forwards:6
          To:MediaServer <sip:mresources@example.com>;tag=62784
          From:Sarvi <sip:sarvi@example.com>;tag=1928301774
          Call-ID:a84b4c76e66710
          CSeq:314162 ACK
          Content-Length:0

                          Example: Add Recognizer

   This example exchange continues from the previous figure and
   deallocates the recognizer channel.  Since a recognizer no longer
   needs to receive an audio stream, this interaction also updates the
   RTP media session to recvonly.

   C->S:  INVITE sip:mresources@server.example.com SIP/2.0
          Via:SIP/2.0/TCP client.atlanta.example.com:5060;
           branch=z9hG4bK74bf5
          Max-Forwards:6



Burnett & Shanmugham         Standards Track                   [Page 20]

RFC 6787                         MRCPv2                    November 2012


          To:MediaServer <sip:mresources@example.com>;tag=62784
          From:sarvi <sip:sarvi@example.com>;tag=1928301774
          Call-ID:a84b4c76e66710
          CSeq:314163 INVITE
          Contact:<sip:sarvi@client.example.com>
          Content-Type:application/sdp
          Content-Length:...

          v=0
          o=sarvi 2890844526 2890844528 IN IP4 192.0.2.12
          s=-
          c=IN IP4 192.0.2.12
          t=0 0
          m=application 9 TCP/MRCPv2 1
          a=resource:speechsynth
          a=cmid:1
          m=audio 49170 RTP/AVP 0
          a=rtpmap:0 pcmu/8000
          a=recvonly
          a=mid:1
          m=application 0 TCP/MRCPv2 1
          a=resource:speechrecog
          a=cmid:1


   S->C:  SIP/2.0 200 OK
          Via:SIP/2.0/TCP client.atlanta.example.com:5060;
           branch=z9hG4bK74bf5;received=192.0.32.10
          To:MediaServer <sip:mresources@example.com>;tag=62784
          From:sarvi <sip:sarvi@example.com>;tag=1928301774
          Call-ID:a84b4c76e66710
          CSeq:314163 INVITE
          Contact:<sip:mresources@server.example.com>
          Content-Type:application/sdp
          Content-Length:...

          v=0
          o=- 2890842808 2890842810 IN IP4 192.0.2.11
          s=-
          c=IN IP4 192.0.2.11
          t=0 0
          m=application 32416 TCP/MRCPv2 1
          a=channel:32AECB234338@speechsynth
          a=cmid:1
          m=audio 48260 RTP/AVP 0
          a=rtpmap:0 pcmu/8000
          a=sendonly
          a=mid:1



Burnett & Shanmugham         Standards Track                   [Page 21]

RFC 6787                         MRCPv2                    November 2012


          m=application 0 TCP/MRCPv2 1
          a=channel:32AECB234338@speechrecog
          a=cmid:1

   C->S:  ACK sip:mresources@server.example.com SIP/2.0
          Via:SIP/2.0/TCP client.atlanta.example.com:5060;
           branch=z9hG4bK74bf6
          Max-Forwards:6
          To:MediaServer <sip:mresources@example.com>;tag=62784
          From:Sarvi <sip:sarvi@example.com>;tag=1928301774
          Call-ID:a84b4c76e66710
          CSeq:314163 ACK
          Content-Length:0

                      Example: Deallocate Recognizer

4.4.  Media Streams and RTP Ports



   Since MRCPv2 resources either generate or consume media streams, the
   client or the server needs to associate media sessions with their
   corresponding resource or resources.  More than one resource could be
   associated with a single media session or each resource could be
   assigned a separate media session.  Also, note that more than one
   media session can be associated with a single resource if need be,
   but this scenario is not useful for the current set of resources.
   For example, a synthesizer and a recognizer could be associated to
   the same media session (m=audio line), if it is opened in "sendrecv"
   mode.  Alternatively, the recognizer could have its own "sendonly"
   audio session, and the synthesizer could have its own "recvonly"
   audio session.

   The association between control channels and their corresponding
   media sessions is established using a new "resource channel media
   identifier" media-level attribute ("cmid").  Valid values of this
   attribute are the values of the "mid" attribute defined in RFC 5888
   [RFC5888].  If there is more than one audio "m=" line, then each
   audio "m=" line MUST have a "mid" attribute.  Each control "m=" line
   MAY have one or more "cmid" attributes that match the resource
   control channel to the "mid" attributes of the audio "m=" lines it is
   associated with.  Note that if a control "m=" line does not have a
   "cmid" attribute it will not be associated with any media.  The
   operations on such a resource will hence be limited.  For example, if
   it was a recognizer resource, the RECOGNIZE method requires an
   associated media to process while the INTERPRET method does not.  The
   formatting of the "cmid" attribute is described by the following
   ABNF:





Burnett & Shanmugham         Standards Track                   [Page 22]

RFC 6787                         MRCPv2                    November 2012


   cmid-attribute     = "a=cmid:" identification-tag
   identification-tag = token

   To allow this flexible mapping of media sessions to MRCPv2 control
   channels, a single audio "m=" line can be associated with multiple
   resources, or each resource can have its own audio "m=" line.  For
   example, if the client wants to allocate a recognizer and a
   synthesizer and associate them with a single two-way audio stream,
   the SDP offer would contain two control "m=" lines and a single audio
   "m=" line with an attribute of "sendrecv".  Each of the control "m="
   lines would have a "cmid" attribute whose value matches the "mid" of
   the audio "m=" line.  If, on the other hand, the client wants to
   allocate a recognizer and a synthesizer each with its own separate
   audio stream, the SDP offer would carry two control "m=" lines (one
   for the recognizer and another for the synthesizer) and two audio
   "m=" lines (one with the attribute "sendonly" and another with
   attribute "recvonly").  The "cmid" attribute of the recognizer
   control "m=" line would match the "mid" value of the "sendonly" audio
   "m=" line, and the "cmid" attribute of the synthesizer control "m="
   line would match the "mid" attribute of the "recvonly" "m=" line.

   When a server receives media (e.g., audio) on a media session that is
   associated with more than one media processing resource, it is the
   responsibility of the server to receive and fork the media to the
   resources that need to consume it.  If multiple resources in an
   MRCPv2 session are generating audio (or other media) to be sent on a
   single associated media session, it is the responsibility of the
   server either to multiplex the multiple streams onto the single RTP
   session or to contain an embedded RTP mixer (see RFC 3550 [RFC3550])
   to combine the multiple streams into one.  In the former case, the
   media stream will contain RTP packets generated by different sources,
   and hence the packets will have different Synchronization Source
   Identifiers (SSRCs).  In the latter case, the RTP packets will
   contain multiple Contributing Source Identifiers (CSRCs)
   corresponding to the original streams before being combined by the
   mixer.  If an MRCPv2 server implementation neither multiplexes nor
   mixes, it MUST disallow the client from associating multiple such
   resources to a single audio stream by rejecting the SDP offer with a
   SIP 488 "Not Acceptable" error.  Note that there is a large installed
   base that will return a SIP 501 "Not Implemented" error in this case.
   To facilitate interoperability with this installed base, new
   implementations SHOULD treat a 501 in this context as a 488 when it
   is received from an element known to be a legacy implementation.








Burnett & Shanmugham         Standards Track                   [Page 23]

RFC 6787                         MRCPv2                    November 2012


4.5.  MRCPv2 Message Transport



   The MRCPv2 messages defined in this document are transported over a
   TCP or TLS connection between the client and the server.  The method
   for setting up this transport connection and the resource control
   channel is discussed in Sections 4.1 and 4.2.  Multiple resource
   control channels between a client and a server that belong to
   different SIP dialogs can share one or more TLS or TCP connections
   between them; the server and client MUST support this mode of
   operation.  Clients and servers MUST use the MRCPv2 channel
   identifier, carried in the Channel-Identifier header field in
   individual MRCPv2 messages, to differentiate MRCPv2 messages from
   different resource channels (see Section 6.2.1 for details).  All
   MRCPv2 servers MUST support TLS.  Servers MAY use TCP without TLS in
   controlled environments (e.g., not in the public Internet) where both
   nodes are inside a protected perimeter, for example, preventing
   access to the MRCP server from remote nodes outside the controlled
   perimeter.  It is up to the client to choose which mode of transport
   it wants to use for an MRCPv2 session.

   Most examples from here on show only the MRCPv2 messages and do not
   show the SIP messages that may have been used to establish the MRCPv2
   control channel.

4.6.  MRCPv2 Session Termination



   If an MRCP client notices that the underlying connection has been
   closed for one of its MRCP channels, and it has not previously
   initiated a re-INVITE to close that channel, it MUST send a BYE to
   close down the SIP dialog and all other MRCP channels.  If an MRCP
   server notices that the underlying connection has been closed for one
   of its MRCP channels, and it has not previously received and accepted
   a re-INVITE closing that channel, then it MUST send a BYE to close
   down the SIP dialog and all other MRCP channels.

5.  MRCPv2 Specification



   Except as otherwise indicated, MRCPv2 messages are Unicode encoded in
   UTF-8 (RFC 3629 [RFC3629]) to allow many different languages to be
   represented.  DEFINE-GRAMMAR (Section 9.8), for example, is one such
   exception, since its body can contain arbitrary XML in arbitrary (but
   specified via XML) encodings.  MRCPv2 also allows message bodies to
   be represented in other character sets (for example, ISO 8859-1
   [ISO.8859-1.1987]) because, in some locales, other character sets are
   already in widespread use.  The MRCPv2 headers (the first line of an
   MRCP message) and header field names use only the US-ASCII subset of
   UTF-8.




Burnett & Shanmugham         Standards Track                   [Page 24]

RFC 6787                         MRCPv2                    November 2012


   Lines are terminated by CRLF (carriage return, then line feed).
   Also, some parameters in the message may contain binary data or a
   record spanning multiple lines.  Such fields have a length value
   associated with the parameter, which indicates the number of octets
   immediately following the parameter.

5.1.  Common Protocol Elements



   The MRCPv2 message set consists of requests from the client to the
   server, responses from the server to the client, and asynchronous
   events from the server to the client.  All these messages consist of
   a start-line, one or more header fields, an empty line (i.e., a line
   with nothing preceding the CRLF) indicating the end of the header
   fields, and an optional message body.

generic-message  =    start-line
                      message-header
                      CRLF
                      [ message-body ]

message-body     =    *OCTET

start-line       =    request-line / response-line / event-line

message-header   =  1*(generic-header / resource-header / generic-field)

resource-header  =    synthesizer-header
                 /    recognizer-header
                 /    recorder-header
                 /    verifier-header

   The message-body contains resource-specific and message-specific
   data.  The actual media types used to carry the data are specified in
   the sections defining the individual messages.  Generic header fields
   are described in Section 6.2.

   If a message contains a message body, the message MUST contain
   content-headers indicating the media type and encoding of the data in
   the message body.

   Request, response and event messages (described in following
   sections) include the version of MRCP that the message conforms to.
   Version compatibility rules follow [H3.1] regarding version ordering,
   compliance requirements, and upgrading of version numbers.  The
   version information is indicated by "MRCP" (as opposed to "HTTP" in
   [H3.1]) or "MRCP/2.0" (as opposed to "HTTP/1.1" in [H3.1]).  To be
   compliant with this specification, clients and servers sending MRCPv2




Burnett & Shanmugham         Standards Track                   [Page 25]

RFC 6787                         MRCPv2                    November 2012


   messages MUST indicate an mrcp-version of "MRCP/2.0".  ABNF
   productions using mrcp-version can be found in Sections 5.2, 5.3, and
   5.5.

   mrcp-version   =    "MRCP" "/" 1*2DIGIT "." 1*2DIGIT

   The message-length field specifies the length of the message in
   octets, including the start-line, and MUST be the second token from
   the beginning of the message.  This is to make the framing and
   parsing of the message simpler to do.  This field specifies the
   length of the message including data that may be encoded into the
   body of the message.  Note that this value MAY be given as a fixed-
   length integer that is zero-padded (with leading zeros) in order to
   eliminate or reduce inefficiency in cases where the message-length
   value would change as a result of the length of the message-length
   token itself.  This value, as with all lengths in MRCP, is to be
   interpreted as a base-10 number.  In particular, leading zeros do not
   indicate that the value is to be interpreted as a base-8 number.

   message-length =    1*19DIGIT

   The following sample MRCP exchange demonstrates proper message-length
   values.  The values for message-length have been removed from all
   other examples in the specification and replaced by '...' to reduce
   confusion in the case of minor message-length computation errors in
   those examples.

   C->S:   MRCP/2.0 877 INTERPRET 543266
           Channel-Identifier:32AECB23433801@speechrecog
           Interpret-Text:may I speak to Andre Roy
           Content-Type:application/srgs+xml
           Content-ID:<request1@form-level.store>
           Content-Length:661

           <?xml version="1.0"?>
           <!-- the default grammar language is US English -->
           <grammar xmlns="http://www.w3.org/2001/06/grammar"
                    xml:lang="en-US" version="1.0" root="request">
           <!-- single language attachment to tokens -->
               <rule id="yes">
                   <one-of>
                       <item xml:lang="fr-CA">oui</item>
                       <item xml:lang="en-US">yes</item>
                   </one-of>
               </rule>






Burnett & Shanmugham         Standards Track                   [Page 26]

RFC 6787                         MRCPv2                    November 2012


           <!-- single language attachment to a rule expansion -->
               <rule id="request">
                   may I speak to
                   <one-of xml:lang="fr-CA">
                       <item>Michel Tremblay</item>
                       <item>Andre Roy</item>
                   </one-of>
               </rule>
           </grammar>

   S->C:   MRCP/2.0 82 543266 200 IN-PROGRESS
           Channel-Identifier:32AECB23433801@speechrecog

   S->C:   MRCP/2.0 634 INTERPRETATION-COMPLETE 543266 200 COMPLETE
           Channel-Identifier:32AECB23433801@speechrecog
           Completion-Cause:000 success
           Content-Type:application/nlsml+xml
           Content-Length:441

           <?xml version="1.0"?>
           <result xmlns="urn:ietf:params:xml:ns:mrcpv2"
                   xmlns:ex="http://www.example.com/example"
                   grammar="session:request1@form-level.store">
               <interpretation>
                   <instance name="Person">
                       <ex:Person>
                           <ex:Name> Andre Roy </ex:Name>
                       </ex:Person>
                   </instance>
                   <input>   may I speak to Andre Roy </input>
               </interpretation>
           </result>

   All MRCPv2 messages, responses and events MUST carry the Channel-
   Identifier header field so the server or client can differentiate
   messages from different control channels that may share the same
   transport connection.

   In the resource-specific header field descriptions in Sections 8-11,
   a header field is disallowed on a method (request, response, or
   event) for that resource unless specifically listed as being allowed.
   Also, the phrasing "This header field MAY occur on method X"
   indicates that the header field is allowed on that method but is not
   required to be used in every instance of that method.







Burnett & Shanmugham         Standards Track                   [Page 27]

RFC 6787                         MRCPv2                    November 2012


5.2.  Request



   An MRCPv2 request consists of a Request line followed by the message
   header section and an optional message body containing data specific
   to the request message.

   The Request message from a client to the server includes within the
   first line the method to be applied, a method tag for that request
   and the version of the protocol in use.

   request-line   =    mrcp-version SP message-length SP method-name
                       SP request-id CRLF

   The mrcp-version field is the MRCP protocol version that is being
   used by the client.

   The message-length field specifies the length of the message,
   including the start-line.

   Details about the mrcp-version and message-length fields are given in
   Section 5.1.

   The method-name field identifies the specific request that the client
   is making to the server.  Each resource supports a subset of the
   MRCPv2 methods.  The subset for each resource is defined in the
   section of the specification for the corresponding resource.

   method-name    =    generic-method
                  /    synthesizer-method
                  /    recognizer-method
                  /    recorder-method
                  /    verifier-method

   The request-id field is a unique identifier representable as an
   unsigned 32-bit integer created by the client and sent to the server.
   Clients MUST utilize monotonically increasing request-ids for
   consecutive requests within an MRCP session.  The request-id space is
   linear (i.e., not mod(32)), so the space does not wrap, and validity
   can be checked with a simple unsigned comparison operation.  The
   client may choose any initial value for its first request, but a
   small integer is RECOMMENDED to avoid exhausting the space in long
   sessions.  If the server receives duplicate or out-of-order requests,
   the server MUST reject the request with a response code of 410.
   Since request-ids are scoped to the MRCP session, they are unique
   across all TCP connections and all resource channels in the session.

   The server resource MUST use the client-assigned identifier in its
   response to the request.  If the request does not complete



Burnett & Shanmugham         Standards Track                   [Page 28]

RFC 6787                         MRCPv2                    November 2012


   synchronously, future asynchronous events associated with this
   request MUST carry the client-assigned request-id.

   request-id     =    1*10DIGIT

5.3.  Response



   After receiving and interpreting the request message for a method,
   the server resource responds with an MRCPv2 response message.  The
   response consists of a response line followed by the message header
   section and an optional message body containing data specific to the
   method.

   response-line  =    mrcp-version SP message-length SP request-id
                       SP status-code SP request-state CRLF

   The mrcp-version field MUST contain the version of the request if
   supported; otherwise, it MUST contain the highest version of MRCP
   supported by the server.

   The message-length field specifies the length of the message,
   including the start-line.

   Details about the mrcp-version and message-length fields are given in
   Section 5.1.

   The request-id used in the response MUST match the one sent in the
   corresponding request message.

   The status-code field is a 3-digit code representing the success or
   failure or other status of the request.

   status-code     =    3DIGIT

   The request-state field indicates if the action initiated by the
   Request is PENDING, IN-PROGRESS, or COMPLETE.  The COMPLETE status
   means that the request was processed to completion and that there
   will be no more events or other messages from that resource to the
   client with that request-id.  The PENDING status means that the
   request has been placed in a queue and will be processed in first-in-
   first-out order.  The IN-PROGRESS status means that the request is
   being processed and is not yet complete.  A PENDING or IN-PROGRESS
   status indicates that further Event messages may be delivered with
   that request-id.

   request-state    =  "COMPLETE"
                    /  "IN-PROGRESS"
                    /  "PENDING"



Burnett & Shanmugham         Standards Track                   [Page 29]

RFC 6787                         MRCPv2                    November 2012


5.4.  Status Codes



   The status codes are classified under the Success (2xx), Client
   Failure (4xx), and Server Failure (5xx) codes.

     +------------+--------------------------------------------------+
     | Code       | Meaning                                          |
     +------------+--------------------------------------------------+
     | 200        | Success                                          |
     | 201        | Success with some optional header fields ignored |
     +------------+--------------------------------------------------+

                               Success (2xx)

   +--------+----------------------------------------------------------+
   | Code   | Meaning                                                  |
   +--------+----------------------------------------------------------+
   | 401    | Method not allowed                                       |
   | 402    | Method not valid in this state                           |
   | 403    | Unsupported header field                                 |
   | 404    | Illegal value for header field. This is the error for a  |
   |        | syntax violation.                                        |
   | 405    | Resource not allocated for this session or does not      |
   |        | exist                                                    |
   | 406    | Mandatory Header Field Missing                           |
   | 407    | Method or Operation Failed (e.g., Grammar compilation    |
   |        | failed in the recognizer. Detailed cause codes might be  |
   |        | available through a resource-specific header.)           |
   | 408    | Unrecognized or unsupported message entity               |
   | 409    | Unsupported Header Field Value. This is a value that is  |
   |        | syntactically legal but exceeds the implementation's     |
   |        | capabilities or expectations.                            |
   | 410    | Non-Monotonic or Out-of-order sequence number in request.|
   | 411-420| Reserved for future assignment                           |
   +--------+----------------------------------------------------------+

                           Client Failure (4xx)

              +------------+--------------------------------+
              | Code       | Meaning                        |
              +------------+--------------------------------+
              | 501        | Server Internal Error          |
              | 502        | Protocol Version not supported |
              | 503        | Reserved for future assignment |
              | 504        | Message too large              |
              +------------+--------------------------------+

                           Server Failure (5xx)



Burnett & Shanmugham         Standards Track                   [Page 30]

RFC 6787                         MRCPv2                    November 2012


5.5.  Events



   The server resource may need to communicate a change in state or the
   occurrence of a certain event to the client.  These messages are used
   when a request does not complete immediately and the response returns
   a status of PENDING or IN-PROGRESS.  The intermediate results and
   events of the request are indicated to the client through the event
   message from the server.  The event message consists of an event
   header line followed by the message header section and an optional
   message body containing data specific to the event message.  The
   header line has the request-id of the corresponding request and
   status value.  The request-state value is COMPLETE if the request is
   done and this was the last event, else it is IN-PROGRESS.

   event-line       =  mrcp-version SP message-length SP event-name
                       SP request-id SP request-state CRLF

   The mrcp-version used here is identical to the one used in the
   Request/Response line and indicates the highest version of MRCP
   running on the server.

   The message-length field specifies the length of the message,
   including the start-line.

   Details about the mrcp-version and message-length fields are given in
   Section 5.1.

   The event-name identifies the nature of the event generated by the
   media resource.  The set of valid event names depends on the resource
   generating it.  See the corresponding resource-specific section of
   the document.

   event-name       =  synthesizer-event
                    /  recognizer-event
                    /  recorder-event
                    /  verifier-event

   The request-id used in the event MUST match the one sent in the
   request that caused this event.

   The request-state indicates whether the Request/Command causing this
   event is complete or still in progress and whether it is the same as
   the one mentioned in Section 5.3.  The final event for a request has
   a COMPLETE status indicating the completion of the request.







Burnett & Shanmugham         Standards Track                   [Page 31]

RFC 6787                         MRCPv2                    November 2012


6.  MRCPv2 Generic Methods, Headers, and Result Structure



   MRCPv2 supports a set of methods and header fields that are common to
   all resources.  These are discussed here; resource-specific methods
   and header fields are discussed in the corresponding resource-
   specific section of the document.

6.1.  Generic Methods



   MRCPv2 supports two generic methods for reading and writing the state
   associated with a resource.

   generic-method      =    "SET-PARAMS"
                       /    "GET-PARAMS"

   These are described in the following subsections.

6.1.1.  SET-PARAMS



   The SET-PARAMS method, from the client to the server, tells the
   MRCPv2 resource to define parameters for the session, such as voice
   characteristics and prosody on synthesizers, recognition timers on
   recognizers, etc.  If the server accepts and sets all parameters, it
   MUST return a response status-code of 200.  If it chooses to ignore
   some optional header fields that can be safely ignored without
   affecting operation of the server, it MUST return 201.

   If one or more of the header fields being sent is incorrect, error
   403, 404, or 409 MUST be returned as follows:

   o  If one or more of the header fields being set has an illegal
      value, the server MUST reject the request with a 404 Illegal Value
      for Header Field.

   o  If one or more of the header fields being set is unsupported for
      the resource, the server MUST reject the request with a 403
      Unsupported Header Field, except as described in the next
      paragraph.

   o  If one or more of the header fields being set has an unsupported
      value, the server MUST reject the request with a 409 Unsupported
      Header Field Value, except as described in the next paragraph.

   If both error 404 and another error have occurred, only error 404
   MUST be returned.  If both errors 403 and 409 have occurred, but not
   error 404, only error 403 MUST be returned.





Burnett & Shanmugham         Standards Track                   [Page 32]

RFC 6787                         MRCPv2                    November 2012


   If error 403, 404, or 409 is returned, the response MUST include the
   bad or unsupported header fields and their values exactly as they
   were sent from the client.  Session parameters modified using
   SET-PARAMS do not override parameters explicitly specified on
   individual requests or requests that are IN-PROGRESS.

   C->S:  MRCP/2.0 ... SET-PARAMS 543256
          Channel-Identifier:32AECB23433802@speechsynth
          Voice-gender:female
          Voice-variant:3

   S->C:  MRCP/2.0 ... 543256 200 COMPLETE
          Channel-Identifier:32AECB23433802@speechsynth

6.1.2.  GET-PARAMS



   The GET-PARAMS method, from the client to the server, asks the MRCPv2
   resource for its current session parameters, such as voice
   characteristics and prosody on synthesizers, recognition timers on
   recognizers, etc.  For every header field the client sends in the
   request without a value, the server MUST include the header field and
   its corresponding value in the response.  If no parameter header
   fields are specified by the client, then the server MUST return all
   the settable parameters and their values in the corresponding header
   section of the response, including vendor-specific parameters.  Such
   wildcard parameter requests can be very processing-intensive, since
   the number of settable parameters can be large depending on the
   implementation.  Hence, it is RECOMMENDED that the client not use the
   wildcard GET-PARAMS operation very often.  Note that GET-PARAMS
   returns header field values that apply to the whole session and not
   values that have a request-level scope.  For example, Input-Waveform-
   URI is a request-level header field and thus would not be returned by
   GET-PARAMS.

   If all of the header fields requested are supported, the server MUST
   return a response status-code of 200.  If some of the header fields
   being retrieved are unsupported for the resource, the server MUST
   reject the request with a 403 Unsupported Header Field.  Such a
   response MUST include the unsupported header fields exactly as they
   were sent from the client, without values.

   C->S:   MRCP/2.0 ... GET-PARAMS 543256
           Channel-Identifier:32AECB23433802@speechsynth
           Voice-gender:
           Voice-variant:
           Vendor-Specific-Parameters:com.example.param1;
                         com.example.param2




Burnett & Shanmugham         Standards Track                   [Page 33]

RFC 6787                         MRCPv2                    November 2012


   S->C:   MRCP/2.0 ... 543256 200 COMPLETE
           Channel-Identifier:32AECB23433802@speechsynth
           Voice-gender:female
           Voice-variant:3
           Vendor-Specific-Parameters:com.example.param1="Company Name";
                         com.example.param2="124324234@example.com"

6.2.  Generic Message Headers



   All MRCPv2 header fields, which include both the generic-headers
   defined in the following subsections and the resource-specific header
   fields defined later, follow the same generic format as that given in
   Section 3.1 of RFC 5322 [RFC5322].  Each header field consists of a
   name followed by a colon (":") and the value.  Header field names are
   case-insensitive.  The value MAY be preceded by any amount of LWS
   (linear white space), though a single SP (space) is preferred.
   Header fields may extend over multiple lines by preceding each extra
   line with at least one SP or HT (horizontal tab).

   generic-field  = field-name ":" [ field-value ]
   field-name     = token
   field-value    = *LWS field-content *( CRLF 1*LWS field-content)
   field-content  = <the OCTETs making up the field-value
                    and consisting of either *TEXT or combinations
                    of token, separators, and quoted-string>

   The field-content does not include any leading or trailing LWS (i.e.,
   linear white space occurring before the first non-whitespace
   character of the field-value or after the last non-whitespace
   character of the field-value).  Such leading or trailing LWS MAY be
   removed without changing the semantics of the field value.  Any LWS
   that occurs between field-content MAY be replaced with a single SP
   before interpreting the field value or forwarding the message
   downstream.

   MRCPv2 servers and clients MUST NOT depend on header field order.  It
   is RECOMMENDED to send general-header fields first, followed by
   request-header or response-header fields, and ending with the entity-
   header fields.  However, MRCPv2 servers and clients MUST be prepared
   to process the header fields in any order.  The only exception to
   this rule is when there are multiple header fields with the same name
   in a message.

   Multiple header fields with the same name MAY be present in a message
   if and only if the entire value for that header field is defined as a
   comma-separated list [i.e., #(values)].





Burnett & Shanmugham         Standards Track                   [Page 34]

RFC 6787                         MRCPv2                    November 2012


   Since vendor-specific parameters may be order-dependent, it MUST be
   possible to combine multiple header fields of the same name into one
   "name:value" pair without changing the semantics of the message, by
   appending each subsequent value to the first, each separated by a
   comma.  The order in which header fields with the same name are
   received is therefore significant to the interpretation of the
   combined header field value, and thus an intermediary MUST NOT change
   the order of these values when a message is forwarded.

   generic-header      =    channel-identifier
                       /    accept
                       /    active-request-id-list
                       /    proxy-sync-id
                       /    accept-charset
                       /    content-type
                       /    content-id
                       /    content-base
                       /    content-encoding
                       /    content-location
                       /    content-length
                       /    fetch-timeout
                       /    cache-control
                       /    logging-tag
                       /    set-cookie
                       /    vendor-specific

6.2.1.  Channel-Identifier



   All MRCPv2 requests, responses, and events MUST contain the Channel-
   Identifier header field.  The value is allocated by the server when a
   control channel is added to the session and communicated to the
   client by the "a=channel" attribute in the SDP answer from the
   server.  The header field value consists of 2 parts separated by the
   '@' symbol.  The first part is an unambiguous string identifying the
   MRCPv2 session.  The second part is a string token that specifies one
   of the media processing resource types listed in Section 3.1.  The
   unambiguous string (first part) MUST be difficult to guess, unique
   among the resource instances managed by the server, and common to all
   resource channels with that server established through a single SIP
   dialog.

   channel-identifier  = "Channel-Identifier" ":" channel-id CRLF
   channel-id          = 1*alphanum "@" 1*alphanum








Burnett & Shanmugham         Standards Track                   [Page 35]

RFC 6787                         MRCPv2                    November 2012


6.2.2.  Accept



   The Accept header field follows the syntax defined in [H14.1].  The
   semantics are also identical, with the exception that if no Accept
   header field is present, the server MUST assume a default value that
   is specific to the resource type that is being controlled.  This
   default value can be changed for a resource on a session by sending
   this header field in a SET-PARAMS method.  The current default value
   of this header field for a resource in a session can be found through
   a GET-PARAMS method.  This header field MAY occur on any request.

6.2.3.  Active-Request-Id-List



   In a request, this header field indicates the list of request-ids to
   which the request applies.  This is useful when there are multiple
   requests that are PENDING or IN-PROGRESS and the client wants this
   request to apply to one or more of these specifically.

   In a response, this header field returns the list of request-ids that
   the method modified or affected.  There could be one or more requests
   in a request-state of PENDING or IN-PROGRESS.  When a method
   affecting one or more PENDING or IN-PROGRESS requests is sent from
   the client to the server, the response MUST contain the list of
   request-ids that were affected or modified by this command in its
   header section.

   The Active-Request-Id-List is only used in requests and responses,
   not in events.

   For example, if a STOP request with no Active-Request-Id-List is sent
   to a synthesizer resource that has one or more SPEAK requests in the
   PENDING or IN-PROGRESS state, all SPEAK requests MUST be cancelled,
   including the one IN-PROGRESS.  The response to the STOP request
   contains in the Active-Request-Id-List value the request-ids of all
   the SPEAK requests that were terminated.  After sending the STOP
   response, the server MUST NOT send any SPEAK-COMPLETE or RECOGNITION-
   COMPLETE events for the terminated requests.

   active-request-id-list  =  "Active-Request-Id-List" ":"
                              request-id *("," request-id) CRLF

6.2.4.  Proxy-Sync-Id



   When any server resource generates a "barge-in-able" event, it also
   generates a unique tag.  The tag is sent as this header field's value
   in an event to the client.  The client then acts as an intermediary
   among the server resources and sends a BARGE-IN-OCCURRED method to
   the synthesizer server resource with the Proxy-Sync-Id it received



Burnett & Shanmugham         Standards Track                   [Page 36]

RFC 6787                         MRCPv2                    November 2012


   from the server resource.  When the recognizer and synthesizer
   resources are part of the same session, they may choose to work
   together to achieve quicker interaction and response.  Here, the
   Proxy-Sync-Id helps the resource receiving the event, intermediated
   by the client, to decide if this event has been processed through a
   direct interaction of the resources.  This header field MAY occur
   only on events and the BARGE-IN-OCCURRED method.  The name of this
   header field contains the word 'proxy' only for historical reasons
   and does not imply that a proxy server is involved.

   proxy-sync-id    =  "Proxy-Sync-Id" ":" 1*VCHAR CRLF

6.2.5.  Accept-Charset



   See [H14.2].  This specifies the acceptable character sets for
   entities returned in the response or events associated with this
   request.  This is useful in specifying the character set to use in
   the Natural Language Semantic Markup Language (NLSML) results of a
   RECOGNITION-COMPLETE event.  This header field is only used on
   requests.

6.2.6.  Content-Type



   See [H14.17].  MRCPv2 supports a restricted set of registered media
   types for content, including speech markup, grammar, and recognition
   results.  The content types applicable to each MRCPv2 resource-type
   are specified in the corresponding section of the document and are
   registered in the MIME Media Types registry maintained by IANA.  The
   multipart content type "multipart/mixed" is supported to communicate
   multiple of the above mentioned contents, in which case the body
   parts MUST NOT contain any MRCPv2-specific header fields.  This
   header field MAY occur on all messages.

   content-type     =    "Content-Type" ":" media-type-value CRLF

   media-type-value =    type "/" subtype *( ";" parameter )

   type             =    token

   subtype          =    token

   parameter        =    attribute "=" value

   attribute        =    token

   value            =    token / quoted-string





Burnett & Shanmugham         Standards Track                   [Page 37]

RFC 6787                         MRCPv2                    November 2012


6.2.7.  Content-ID



   This header field contains an ID or name for the content by which it
   can be referenced.  This header field operates according to the
   specification in RFC 2392 [RFC2392] and is required for content
   disambiguation in multipart messages.  In MRCPv2, whenever the
   associated content is stored by either the client or the server, it
   MUST be retrievable using this ID.  Such content can be referenced
   later in a session by addressing it with the 'session' URI scheme
   described in Section 13.6.  This header field MAY occur on all
   messages.

6.2.8.  Content-Base



   The Content-Base entity-header MAY be used to specify the base URI
   for resolving relative URIs within the entity.

   content-base      = "Content-Base" ":" absoluteURI CRLF

   Note, however, that the base URI of the contents within the entity-
   body may be redefined within that entity-body.  An example of this
   would be multipart media, which in turn can have multiple entities
   within it.  This header field MAY occur on all messages.

6.2.9.  Content-Encoding



   The Content-Encoding entity-header is used as a modifier to the
   Content-Type.  When present, its value indicates what additional
   content encoding has been applied to the entity-body, and thus what
   decoding mechanisms must be applied in order to obtain the Media Type
   referenced by the Content-Type header field.  Content-Encoding is
   primarily used to allow a document to be compressed without losing
   the identity of its underlying media type.  Note that the SIP session
   can be used to determine accepted encodings (see Section 7).  This
   header field MAY occur on all messages.

   content-encoding  = "Content-Encoding" ":"
                       *WSP content-coding
                       *(*WSP "," *WSP content-coding *WSP )
                       CRLF

   Content codings are defined in [H3.5].  An example of its use is
   Content-Encoding:gzip

   If multiple encodings have been applied to an entity, the content
   encodings MUST be listed in the order in which they were applied.





Burnett & Shanmugham         Standards Track                   [Page 38]

RFC 6787                         MRCPv2                    November 2012


6.2.10.  Content-Location



   The Content-Location entity-header MAY be used to supply the resource
   location for the entity enclosed in the message when that entity is
   accessible from a location separate from the requested resource's
   URI.  Refer to [H14.14].

   content-location  =  "Content-Location" ":"
                        ( absoluteURI / relativeURI ) CRLF

   The Content-Location value is a statement of the location of the
   resource corresponding to this particular entity at the time of the
   request.  This header field is provided for optimization purposes
   only.  The receiver of this header field MAY assume that the entity
   being sent is identical to what would have been retrieved or might
   already have been retrieved from the Content-Location URI.

   For example, if the client provided a grammar markup inline, and it
   had previously retrieved it from a certain URI, that URI can be
   provided as part of the entity, using the Content-Location header
   field.  This allows a resource like the recognizer to look into its
   cache to see if this grammar was previously retrieved, compiled, and
   cached.  In this case, it might optimize by using the previously
   compiled grammar object.

   If the Content-Location is a relative URI, the relative URI is
   interpreted relative to the Content-Base URI.  This header field MAY
   occur on all messages.

6.2.11.  Content-Length



   This header field contains the length of the content of the message
   body (i.e., after the double CRLF following the last header field).
   Unlike in HTTP, it MUST be included in all messages that carry
   content beyond the header section.  If it is missing, a default value
   of zero is assumed.  Otherwise, it is interpreted according to
   [H14.13].  When a message having no use for a message body contains
   one, i.e., the Content-Length is non-zero, the receiver MUST ignore
   the content of the message body.  This header field MAY occur on all
   messages.

   content-length  =  "Content-Length" ":" 1*19DIGIT CRLF

6.2.12.  Fetch Timeout



   When the recognizer or synthesizer needs to fetch documents or other
   resources, this header field controls the corresponding URI access
   properties.  This defines the timeout for content that the server may



Burnett & Shanmugham         Standards Track                   [Page 39]

RFC 6787                         MRCPv2                    November 2012


   need to fetch over the network.  The value is interpreted to be in
   milliseconds and ranges from 0 to an implementation-specific maximum
   value.  It is RECOMMENDED that servers be cautious about accepting
   long timeout values.  The default value for this header field is
   implementation specific.  This header field MAY occur in DEFINE-
   GRAMMAR, RECOGNIZE, SPEAK, SET-PARAMS, or GET-PARAMS.

   fetch-timeout       =   "Fetch-Timeout" ":" 1*19DIGIT CRLF

6.2.13.  Cache-Control



   If the server implements content caching, it MUST adhere to the cache
   correctness rules of HTTP 1.1 [RFC2616] when accessing and caching
   stored content.  In particular, the "expires" and "cache-control"
   header fields of the cached URI or document MUST be honored and take
   precedence over the Cache-Control defaults set by this header field.
   The Cache-Control directives are used to define the default caching
   algorithms on the server for the session or request.  The scope of
   the directive is based on the method it is sent on.  If the directive
   is sent on a SET-PARAMS method, it applies for all requests for
   external documents the server makes during that session, unless it is
   overridden by a Cache-Control header field on an individual request.
   If the directives are sent on any other requests, they apply only to
   external document requests the server makes for that request.  An
   empty Cache-Control header field on the GET-PARAMS method is a
   request for the server to return the current Cache-Control directives
   setting on the server.  This header field MAY occur only on requests.

   cache-control    =    "Cache-Control" ":"
                         [*WSP cache-directive
                         *( *WSP "," *WSP cache-directive *WSP )]
                         CRLF

   cache-directive     = "max-age" "=" delta-seconds
                       / "max-stale" [ "=" delta-seconds ]
                       / "min-fresh" "=" delta-seconds

   delta-seconds       = 1*19DIGIT

   Here, delta-seconds is a decimal time value specifying the number of
   seconds since the instant the message response or data was received
   by the server.

   The different cache-directive options allow the client to ask the
   server to override the default cache expiration mechanisms:






Burnett & Shanmugham         Standards Track                   [Page 40]

RFC 6787                         MRCPv2                    November 2012


   max-age        Indicates that the client can tolerate the server
                  using content whose age is no greater than the
                  specified time in seconds.  Unless a "max-stale"
                  directive is also included, the client is not willing
                  to accept a response based on stale data.

   min-fresh      Indicates that the client is willing to accept a
                  server response with cached data whose expiration is
                  no less than its current age plus the specified time
                  in seconds.  If the server's cache time-to-live
                  exceeds the client-supplied min-fresh value, the
                  server MUST NOT utilize cached content.

   max-stale      Indicates that the client is willing to allow a server
                  to utilize cached data that has exceeded its
                  expiration time.  If "max-stale" is assigned a value,
                  then the client is willing to allow the server to use
                  cached data that has exceeded its expiration time by
                  no more than the specified number of seconds.  If no
                  value is assigned to "max-stale", then the client is
                  willing to allow the server to use stale data of any
                  age.

   If the server cache is requested to use stale response/data without
   validation, it MAY do so only if this does not conflict with any
   "MUST"-level requirements concerning cache validation (e.g., a "must-
   revalidate" Cache-Control directive in the HTTP 1.1 specification
   pertaining to the corresponding URI).

   If both the MRCPv2 Cache-Control directive and the cached entry on
   the server include "max-age" directives, then the lesser of the two
   values is used for determining the freshness of the cached entry for
   that request.

6.2.14.  Logging-Tag



   This header field MAY be sent as part of a SET-PARAMS/GET-PARAMS
   method to set or retrieve the logging tag for logs generated by the
   server.  Once set, the value persists until a new value is set or the
   session ends.  The MRCPv2 server MAY provide a mechanism to create
   subsets of its output logs so that system administrators can examine
   or extract only the log file portion during which the logging tag was
   set to a certain value.

   It is RECOMMENDED that clients include in the logging tag information
   to identify the MRCPv2 client User Agent, so that one can determine
   which MRCPv2 client request generated a given log message at the
   server.  It is also RECOMMENDED that MRCPv2 clients not log



Burnett & Shanmugham         Standards Track                   [Page 41]

RFC 6787                         MRCPv2                    November 2012


   personally identifiable information such as credit card numbers and
   national identification numbers.

   logging-tag    = "Logging-Tag" ":" 1*UTFCHAR CRLF

6.2.15.  Set-Cookie



   Since the associated HTTP client on an MRCPv2 server fetches
   documents for processing on behalf of the MRCPv2 client, the cookie
   store in the HTTP client of the MRCPv2 server is treated as an
   extension of the cookie store in the HTTP client of the MRCPv2
   client.  This requires that the MRCPv2 client and server be able to
   synchronize their common cookie store as needed.  To enable the
   MRCPv2 client to push its stored cookies to the MRCPv2 server and get
   new cookies from the MRCPv2 server stored back to the MRCPv2 client,
   the Set-Cookie entity-header field MAY be included in MRCPv2 requests
   to update the cookie store on a server and be returned in final
   MRCPv2 responses or events to subsequently update the client's own
   cookie store.  The stored cookies on the server persist for the
   duration of the MRCPv2 session and MUST be destroyed at the end of
   the session.  To ensure support for cookies, MRCPv2 clients and
   servers MUST support the Set-Cookie entity-header field.

   Note that it is the MRCPv2 client that determines which, if any,
   cookies are sent to the server.  There is no requirement that all
   cookies be shared.  Rather, it is RECOMMENDED that MRCPv2 clients
   communicate only cookies needed by the MRCPv2 server to process its
   requests.

 set-cookie      =       "Set-Cookie:" cookies CRLF
 cookies         =       cookie *("," *LWS cookie)
 cookie          =       attribute "=" value *(";" cookie-av)
 cookie-av       =       "Comment" "=" value
                 /       "Domain" "=" value
                 /       "Max-Age" "=" value
                 /       "Path" "=" value
                 /       "Secure"
                 /       "Version" "=" 1*19DIGIT
                 /       "Age" "=" delta-seconds

 set-cookie        = "Set-Cookie:" SP set-cookie-string
 set-cookie-string = cookie-pair *( ";" SP cookie-av )
 cookie-pair       = cookie-name "=" cookie-value
 cookie-name       = token
 cookie-value      = *cookie-octet / ( DQUOTE *cookie-octet DQUOTE )
 cookie-octet      = %x21 / %x23-2B / %x2D-3A / %x3C-5B / %x5D-7E
 token             = <token, defined in [RFC2616], Section 2.2>




Burnett & Shanmugham         Standards Track                   [Page 42]

RFC 6787                         MRCPv2                    November 2012


 cookie-av         = expires-av / max-age-av / domain-av /
                      path-av / secure-av / httponly-av /
                      extension-av / age-av
 expires-av        = "Expires=" sane-cookie-date
 sane-cookie-date  = <rfc1123-date, defined in [RFC2616], Section 3.3.1>
 max-age-av        = "Max-Age=" non-zero-digit *DIGIT
 non-zero-digit    = %x31-39
 domain-av         = "Domain=" domain-value
 domain-value      = <subdomain>
 path-av           = "Path=" path-value
 path-value        = <any CHAR except CTLs or ";">
 secure-av         = "Secure"
 httponly-av       = "HttpOnly"
 extension-av      = <any CHAR except CTLs or ";">
 age-av            = "Age=" delta-seconds

   The Set-Cookie header field is specified in RFC 6265 [RFC6265].  The
   "Age" attribute is introduced in this specification to indicate the
   age of the cookie and is OPTIONAL.  An MRCPv2 client or server MUST
   calculate the age of the cookie according to the age calculation
   rules in the HTTP/1.1 specification [RFC2616] and append the "Age"
   attribute accordingly.  This attribute is provided because time may
   have passed since the client received the cookie from an HTTP server.
   Rather than having the client reduce Max-Age by the actual age, it
   passes Max-Age verbatim and appends the "Age" attribute, thus
   maintaining the cookie as received while still accounting for the
   fact that time has passed.

   The MRCPv2 client or server MUST supply defaults for the "Domain" and
   "Path" attributes, as specified in RFC 6265, if they are omitted by
   the HTTP origin server.  Note that there is no leading dot present in
   the "Domain" attribute value in this case.  Although an explicitly
   specified "Domain" value received via the HTTP protocol may be
   modified to include a leading dot, an MRCPv2 client or server MUST
   NOT
modify the "Domain" value when received via the MRCPv2 protocol.

   An MRCPv2 client or server MAY combine multiple cookie header fields
   of the same type into a single "field-name:field-value" pair as
   described in Section 6.2.

   The Set-Cookie header field MAY be specified in any request that
   subsequently results in the server performing an HTTP access.  When a
   server receives new cookie information from an HTTP origin server,
   and assuming the cookie store is modified according to RFC 6265, the
   server MUST return the new cookie information in the MRCPv2 COMPLETE
   response or event, as appropriate, to allow the client to update its
   own cookie store.




Burnett & Shanmugham         Standards Track                   [Page 43]

RFC 6787                         MRCPv2                    November 2012


   The SET-PARAMS request MAY specify the Set-Cookie header field to
   update the cookie store on a server.  The GET-PARAMS request MAY be
   used to return the entire cookie store of "Set-Cookie" type cookies
   to the client.

6.2.16.  Vendor-Specific Parameters



   This set of header fields allows for the client to set or retrieve
   vendor-specific parameters.

   vendor-specific          =    "Vendor-Specific-Parameters" ":"
                                 [vendor-specific-av-pair
                                 *(";" vendor-specific-av-pair)] CRLF

   vendor-specific-av-pair  = vendor-av-pair-name "="
                              value

   vendor-av-pair-name     = 1*UTFCHAR

   Header fields of this form MAY be sent in any method (request) and
   are used to manage implementation-specific parameters on the server
   side.  The vendor-av-pair-name follows the reverse Internet Domain
   Name convention (see Section 13.1.6 for syntax and registration
   information).  The value of the vendor attribute is specified after
   the "=" symbol and MAY be quoted.  For example:

   com.example.companyA.paramxyz=256
   com.example.companyA.paramabc=High
   com.example.companyB.paramxyz=Low

   When used in GET-PARAMS to get the current value of these parameters
   from the server, this header field value MAY contain a semicolon-
   separated list of implementation-specific attribute names.

6.3.  Generic Result Structure



   Result data from the server for the Recognizer and Verifier resources
   is carried as a typed media entity in the MRCPv2 message body of
   various events.  The Natural Language Semantics Markup Language
   (NLSML), an XML markup based on an early draft from the W3C, is the
   default standard for returning results back to the client.  Hence,
   all servers implementing these resource types MUST support the media
   type 'application/nlsml+xml'.  The Extensible MultiModal Annotation
   (EMMA) [W3C.REC-emma-20090210] format can be used to return results
   as well.  This can be done by negotiating the format at session
   establishment time with SDP (a=resultformat:application/emma+xml) or
   with SIP (Allow/Accept).  With SIP, for example, if a client wants




Burnett & Shanmugham         Standards Track                   [Page 44]

RFC 6787                         MRCPv2                    November 2012


   results in EMMA, an MRCPv2 server can route the request to another
   server that supports EMMA by inspecting the SIP header fields, rather
   than having to inspect the SDP.

   MRCPv2 uses this representation to convey content among the clients
   and servers that generate and make use of the markup.  MRCPv2 uses
   NSLML specifically to convey recognition, enrollment, and
   verification results between the corresponding resource on the MRCPv2
   server and the MRCPv2 client.  Details of this result format are
   fully described in Section 6.3.1.

   Content-Type:application/nlsml+xml
   Content-Length:...

   <?xml version="1.0"?>
   <result xmlns="urn:ietf:params:xml:ns:mrcpv2"
           xmlns:ex="http://www.example.com/example"
           grammar="http://theYesNoGrammar">
       <interpretation>
           <instance>
                   <ex:response>yes</ex:response>
           </instance>
           <input>OK</input>
       </interpretation>
   </result>

                              Result Example

6.3.1.  Natural Language Semantics Markup Language



   The Natural Language Semantics Markup Language (NLSML) is an XML data
   structure with elements and attributes designed to carry result
   information from recognizer (including enrollment) and verifier
   resources.  The normative definition of NLSML is the RelaxNG schema
   in Section 16.1.  Note that the elements and attributes of this
   format are defined in the MRCPv2 namespace.  In the result structure,
   they must either be prefixed by a namespace prefix declared within
   the result or must be children of an element identified as belonging
   to the respective namespace.  For details on how to use XML
   Namespaces, see [W3C.REC-xml-names11-20040204].  Section 2 of
   [W3C.REC-xml-names11-20040204] provides details on how to declare
   namespaces and namespace prefixes.

   The root element of NLSML is <result>.  Optional child elements are
   <interpretation>, <enrollment-result>, and <verification-result>, at
   least one of which must be present.  A single <result> MAY contain
   any or all of the optional child elements.  Details of the <result>
   and <interpretation> elements and their subelements and attributes



Burnett & Shanmugham         Standards Track                   [Page 45]

RFC 6787                         MRCPv2                    November 2012


   can be found in Section 9.6.  Details of the <enrollment-result>
   element and its subelements can be found in Section 9.7.  Details of
   the <verification-result> element and its subelements can be found in
   Section 11.5.2.

7.  Resource Discovery



   Server resources may be discovered and their capabilities learned by
   clients through standard SIP machinery.  The client MAY issue a SIP
   OPTIONS transaction to a server, which has the effect of requesting
   the capabilities of the server.  The server MUST respond to such a
   request with an SDP-encoded description of its capabilities according
   to RFC 3264 [RFC3264].  The MRCPv2 capabilities are described by a
   single "m=" line containing the media type "application" and
   transport type "TCP/TLS/MRCPv2" or "TCP/MRCPv2".  There MUST be one
   "resource" attribute for each media resource that the server
   supports, and it has the resource type identifier as its value.

   The SDP description MUST also contain "m=" lines describing the audio
   capabilities and the coders the server supports.

   In this example, the client uses the SIP OPTIONS method to query the
   capabilities of the MRCPv2 server.

   C->S:
        OPTIONS sip:mrcp@server.example.com SIP/2.0
        Via:SIP/2.0/TCP client.atlanta.example.com:5060;
         branch=z9hG4bK74bf7
        Max-Forwards:6
        To:<sip:mrcp@example.com>
        From:Sarvi <sip:sarvi@example.com>;tag=1928301774
        Call-ID:a84b4c76e66710
        CSeq:63104 OPTIONS
        Contact:<sip:sarvi@client.example.com>
        Accept:application/sdp
        Content-Length:0


   S->C:
        SIP/2.0 200 OK
        Via:SIP/2.0/TCP client.atlanta.example.com:5060;
         branch=z9hG4bK74bf7;received=192.0.32.10
        To:<sip:mrcp@example.com>;tag=62784
        From:Sarvi <sip:sarvi@example.com>;tag=1928301774
        Call-ID:a84b4c76e66710
        CSeq:63104 OPTIONS
        Contact:<sip:mrcp@server.example.com>
        Allow:INVITE, ACK, CANCEL, OPTIONS, BYE



Burnett & Shanmugham         Standards Track                   [Page 46]

RFC 6787                         MRCPv2                    November 2012


        Accept:application/sdp
        Accept-Encoding:gzip
        Accept-Language:en
        Supported:foo
        Content-Type:application/sdp
        Content-Length:...

        v=0
        o=sarvi 2890844536 2890842811 IN IP4 192.0.2.12
        s=-
        i=MRCPv2 server capabilities
        c=IN IP4 192.0.2.12/127
        t=0 0
        m=application 0 TCP/TLS/MRCPv2 1
        a=resource:speechsynth
        a=resource:speechrecog
        a=resource:speakverify
        m=audio 0 RTP/AVP 0 3
        a=rtpmap:0 PCMU/8000
        a=rtpmap:3 GSM/8000

         Using SIP OPTIONS for MRCPv2 Server Capability Discovery

8.  Speech Synthesizer Resource



   This resource processes text markup provided by the client and
   generates a stream of synthesized speech in real time.  Depending
   upon the server implementation and capability of this resource, the
   client can also dictate parameters of the synthesized speech such as
   voice characteristics, speaker speed, etc.

   The synthesizer resource is controlled by MRCPv2 requests from the
   client.  Similarly, the resource can respond to these requests or
   generate asynchronous events to the client to indicate conditions of
   interest to the client during the generation of the synthesized
   speech stream.

   This section applies for the following resource types:

   o  speechsynth

   o  basicsynth

   The capabilities of these resources are defined in Section 3.1.







Burnett & Shanmugham         Standards Track                   [Page 47]

RFC 6787                         MRCPv2                    November 2012


8.1.  Synthesizer State Machine



   The synthesizer maintains a state machine to process MRCPv2 requests
   from the client.  The state transitions shown below describe the
   states of the synthesizer and reflect the state of the request at the
   head of the synthesizer resource queue.  A SPEAK request in the
   PENDING state can be deleted or stopped by a STOP request without
   affecting the state of the resource.

   Idle                    Speaking                  Paused
   State                   State                     State
     |                        |                          |
     |----------SPEAK-------->|                 |--------|
     |<------STOP-------------|             CONTROL      |
     |<----SPEAK-COMPLETE-----|                 |------->|
     |<----BARGE-IN-OCCURRED--|                          |
     |              |---------|                          |
     |          CONTROL       |-----------PAUSE--------->|
     |              |-------->|<----------RESUME---------|
     |                        |               |----------|
     |----------|             |              PAUSE       |
     |    BARGE-IN-OCCURRED   |               |--------->|
     |<---------|             |----------|               |
     |                        |      SPEECH-MARKER       |
     |                        |<---------|               |
     |----------|             |----------|               |
     |         STOP           |       RESUME             |
     |          |             |<---------|               |
     |<---------|             |                          |
     |<---------------------STOP-------------------------|
     |----------|             |                          |
     |     DEFINE-LEXICON     |                          |
     |          |             |                          |
     |<---------|             |                          |
     |<---------------BARGE-IN-OCCURRED------------------|

                         Synthesizer State Machine

8.2.  Synthesizer Methods



   The synthesizer supports the following methods.










Burnett & Shanmugham         Standards Track                   [Page 48]

RFC 6787                         MRCPv2                    November 2012


   synthesizer-method   =  "SPEAK"
                        /  "STOP"
                        /  "PAUSE"
                        /  "RESUME"
                        /  "BARGE-IN-OCCURRED"
                        /  "CONTROL"
                        /  "DEFINE-LEXICON"

8.3.  Synthesizer Events



   The synthesizer can generate the following events.

   synthesizer-event    =  "SPEECH-MARKER"
                        /  "SPEAK-COMPLETE"

8.4.  Synthesizer Header Fields



   A synthesizer method can contain header fields containing request
   options and information to augment the Request, Response, or Event it
   is associated with.

   synthesizer-header  =  jump-size
                       /  kill-on-barge-in
                       /  speaker-profile
                       /  completion-cause
                       /  completion-reason
                       /  voice-parameter
                       /  prosody-parameter
                       /  speech-marker
                       /  speech-language
                       /  fetch-hint
                       /  audio-fetch-hint
                       /  failed-uri
                       /  failed-uri-cause
                       /  speak-restart
                       /  speak-length
                       /  load-lexicon
                       /  lexicon-search-order

8.4.1.  Jump-Size



   This header field MAY be specified in a CONTROL method and controls
   the amount to jump forward or backward in an active SPEAK request.  A
   '+' or '-' indicates a relative value to what is being currently
   played.  This header field MAY also be specified in a SPEAK request
   as a desired offset into the synthesized speech.  In this case, the
   synthesizer MUST begin speaking from this amount of time into the
   speech markup.  Note that an offset that extends beyond the end of



Burnett & Shanmugham         Standards Track                   [Page 49]

RFC 6787                         MRCPv2                    November 2012


   the produced speech will result in audio of length zero.  The
   different speech length units supported are dependent on the
   synthesizer implementation.  If the synthesizer resource does not
   support a unit for the operation, the resource MUST respond with a
   status-code of 409 "Unsupported Header Field Value".

   jump-size             =   "Jump-Size" ":" speech-length-value CRLF

   speech-length-value   =   numeric-speech-length
                         /   text-speech-length

   text-speech-length    =   1*UTFCHAR SP "Tag"

   numeric-speech-length =    ("+" / "-") positive-speech-length

   positive-speech-length =   1*19DIGIT SP numeric-speech-unit

   numeric-speech-unit   =   "Second"
                         /   "Word"
                         /   "Sentence"
                         /   "Paragraph"

8.4.2.  Kill-On-Barge-In



   This header field MAY be sent as part of the SPEAK method to enable
   "kill-on-barge-in" support.  If enabled, the SPEAK method is
   interrupted by DTMF input detected by a signal detector resource or
   by the start of speech sensed or recognized by the speech recognizer
   resource.

   kill-on-barge-in      =   "Kill-On-Barge-In" ":" BOOLEAN CRLF

   The client MUST send a BARGE-IN-OCCURRED method to the synthesizer
   resource when it receives a barge-in-able event from any source.
   This source could be a synthesizer resource or signal detector
   resource and MAY be either local or distributed.  If this header
   field is not specified in a SPEAK request or explicitly set by a
   SET-PARAMS, the default value for this header field is "true".

   If the recognizer or signal detector resource is on the same server
   as the synthesizer and both are part of the same session, the server
   MAY work with both to provide internal notification to the
   synthesizer so that audio may be stopped without having to wait for
   the client's BARGE-IN-OCCURRED event.

   It is generally RECOMMENDED when playing a prompt to the user with
   Kill-On-Barge-In and asking for input, that the client issue the
   RECOGNIZE request ahead of the SPEAK request for optimum performance



Burnett & Shanmugham         Standards Track                   [Page 50]

RFC 6787                         MRCPv2                    November 2012


   and user experience.  This way, it is guaranteed that the recognizer
   is online before the prompt starts playing and the user's speech will
   not be truncated at the beginning (especially for power users).

8.4.3.  Speaker-Profile



   This header field MAY be part of the SET-PARAMS/GET-PARAMS or SPEAK
   request from the client to the server and specifies a URI that
   references the profile of the speaker.  Speaker profiles are
   collections of voice parameters like gender, accent, etc.

   speaker-profile       =   "Speaker-Profile" ":" uri CRLF

8.4.4.  Completion-Cause



   This header field MUST be specified in a SPEAK-COMPLETE event coming
   from the synthesizer resource to the client.  This indicates the
   reason the SPEAK request completed.

   completion-cause      =   "Completion-Cause" ":" 3DIGIT SP
                             1*VCHAR CRLF

   +------------+-----------------------+------------------------------+
   | Cause-Code | Cause-Name            | Description                  |
   +------------+-----------------------+------------------------------+
   | 000        | normal                | SPEAK completed normally.    |
   | 001        | barge-in              | SPEAK request was terminated |
   |            |                       | because of barge-in.         |
   | 002        | parse-failure         | SPEAK request terminated     |
   |            |                       | because of a failure to      |
   |            |                       | parse the speech markup      |
   |            |                       | text.                        |
   | 003        | uri-failure           | SPEAK request terminated     |
   |            |                       | because access to one of the |
   |            |                       | URIs failed.                 |
   | 004        | error                 | SPEAK request terminated     |
   |            |                       | prematurely due to           |
   |            |                       | synthesizer error.           |
   | 005        | language-unsupported  | Language not supported.      |
   | 006        | lexicon-load-failure  | Lexicon loading failed.      |
   | 007        | cancelled             | A prior SPEAK request failed |
   |            |                       | while this one was still in  |
   |            |                       | the queue.                   |
   +------------+-----------------------+------------------------------+

                Synthesizer Resource Completion Cause Codes





Burnett & Shanmugham         Standards Track                   [Page 51]

RFC 6787                         MRCPv2                    November 2012


8.4.5.  Completion-Reason



   This header field MAY be specified in a SPEAK-COMPLETE event coming
   from the synthesizer resource to the client.  This contains the
   reason text behind the SPEAK request completion.  This header field
   communicates text describing the reason for the failure, such as an
   error in parsing the speech markup text.

   completion-reason   =   "Completion-Reason" ":"
                           quoted-string CRLF

   The completion reason text is provided for client use in logs and for
   debugging and instrumentation purposes.  Clients MUST NOT interpret
   the completion reason text.

8.4.6.  Voice-Parameter



   This set of header fields defines the voice of the speaker.

   voice-parameter    =   voice-gender
                       /   voice-age
                       /   voice-variant
                       /   voice-name

   voice-gender        =   "Voice-Gender:" voice-gender-value CRLF
   voice-gender-value  =   "male"
                       /   "female"
                       /   "neutral"
   voice-age           =   "Voice-Age:" 1*3DIGIT CRLF
   voice-variant       =   "Voice-Variant:" 1*19DIGIT CRLF
   voice-name          =   "Voice-Name:"
                           1*UTFCHAR *(1*WSP 1*UTFCHAR) CRLF

   The "Voice-" parameters are derived from the similarly named
   attributes of the voice element specified in W3C's Speech Synthesis
   Markup Language Specification (SSML)
   [W3C.REC-speech-synthesis-20040907].  Legal values for these
   parameters are as defined in that specification.

   These header fields MAY be sent in SET-PARAMS or GET-PARAMS requests
   to define or get default values for the entire session or MAY be sent
   in the SPEAK request to define default values for that SPEAK request.
   Note that SSML content can itself set these values internal to the
   SSML document, of course.







Burnett & Shanmugham         Standards Track                   [Page 52]

RFC 6787                         MRCPv2                    November 2012


   Voice parameter header fields MAY also be sent in a CONTROL method to
   affect a SPEAK request in progress and change its behavior on the
   fly.  If the synthesizer resource does not support this operation, it
   MUST reject the request with a status-code of 403 "Unsupported Header
   Field".

8.4.7.  Prosody-Parameters



   This set of header fields defines the prosody of the speech.

   prosody-parameter   =   "Prosody-" prosody-param-name ":"
                           prosody-param-value CRLF

   prosody-param-name    =    1*VCHAR

   prosody-param-value   =    1*VCHAR

   prosody-param-name is any one of the attribute names under the
   prosody element specified in W3C's Speech Synthesis Markup Language
   Specification [W3C.REC-speech-synthesis-20040907].  The prosody-
   param-value is any one of the value choices of the corresponding
   prosody element attribute from that specification.

   These header fields MAY be sent in SET-PARAMS or GET-PARAMS requests
   to define or get default values for the entire session or MAY be sent
   in the SPEAK request to define default values for that SPEAK request.
   Furthermore, these attributes can be part of the speech text marked
   up in SSML.

   The prosody parameter header fields in the SET-PARAMS or SPEAK
   request only apply if the speech data is of type 'text/plain' and
   does not use a speech markup format.

   These prosody parameter header fields MAY also be sent in a CONTROL
   method to affect a SPEAK request in progress and change its behavior
   on the fly.  If the synthesizer resource does not support this
   operation, it MUST respond back to the client with a status-code of
   403 "Unsupported Header Field".

8.4.8.  Speech-Marker



   This header field contains timestamp information in a "timestamp"
   field.  This is a Network Time Protocol (NTP) [RFC5905] timestamp, a
   64-bit number in decimal form.  It MUST be synced with the Real-Time
   Protocol (RTP) [RFC3550] timestamp of the media stream through the
   Real-Time Control Protocol (RTCP) [RFC3550].





Burnett & Shanmugham         Standards Track                   [Page 53]

RFC 6787                         MRCPv2                    November 2012


   Markers are bookmarks that are defined within the markup.  Most
   speech markup formats provide mechanisms to embed marker fields
   within speech texts.  The synthesizer generates SPEECH-MARKER events
   when it reaches these marker fields.  This header field MUST be part
   of the SPEECH-MARKER event and contain the marker tag value after the
   timestamp, separated by a semicolon.  In these events, the timestamp
   marks the time the text corresponding to the marker was emitted as
   speech by the synthesizer.

   This header field MUST also be returned in responses to STOP,
   CONTROL, and BARGE-IN-OCCURRED methods, in the SPEAK-COMPLETE event,
   and in an IN-PROGRESS SPEAK response.  In these messages, if any
   markers have been encountered for the current SPEAK, the marker tag
   value MUST be the last embedded marker encountered.  If no markers
   have yet been encountered for the current SPEAK, only the timestamp
   is REQUIRED.  Note that in these events, the purpose of this header
   field is to provide timestamp information associated with important
   events within the lifecycle of a request (start of SPEAK processing,
   end of SPEAK processing, receipt of CONTROL/STOP/BARGE-IN-OCCURRED).

   timestamp           =   "timestamp" "=" time-stamp-value

   time-stamp-value    =   1*20DIGIT

   speech-marker       =   "Speech-Marker" ":"
                           timestamp
                           [";" 1*(UTFCHAR / %x20)] CRLF

8.4.9.  Speech-Language



   This header field specifies the default language of the speech data
   if the language is not specified in the markup.  The value of this
   header field MUST follow RFC 5646 [RFC5646] for its values.  The
   header field MAY occur in SPEAK, SET-PARAMS, or GET-PARAMS requests.

   speech-language     =   "Speech-Language" ":" 1*VCHAR CRLF

8.4.10.  Fetch-Hint



   When the synthesizer needs to fetch documents or other resources like
   speech markup or audio files, this header field controls the
   corresponding URI access properties.  This provides client policy on
   when the synthesizer should retrieve content from the server.  A
   value of "prefetch" indicates the content MAY be downloaded when the
   request is received, whereas "safe" indicates that content MUST NOT






Burnett & Shanmugham         Standards Track                   [Page 54]

RFC 6787                         MRCPv2                    November 2012


   be downloaded until actually referenced.  The default value is
   "prefetch".  This header field MAY occur in SPEAK, SET-PARAMS, or
   GET-PARAMS requests.

   fetch-hint          =   "Fetch-Hint" ":" ("prefetch" / "safe") CRLF

8.4.11.  Audio-Fetch-Hint



   When the synthesizer needs to fetch documents or other resources like
   speech audio files, this header field controls the corresponding URI
   access properties.  This provides client policy whether or not the
   synthesizer is permitted to attempt to optimize speech by pre-
   fetching audio.  The value is either "safe" to say that audio is only
   fetched when it is referenced, never before; "prefetch" to permit,
   but not require the implementation to pre-fetch the audio; or
   "stream" to allow it to stream the audio fetches.  The default value
   is "prefetch".  This header field MAY occur in SPEAK, SET-PARAMS, or
   GET-PARAMS requests.

   audio-fetch-hint    =   "Audio-Fetch-Hint" ":"
                           ("prefetch" / "safe" / "stream") CRLF

8.4.12.  Failed-URI



   When a synthesizer method needs a synthesizer to fetch or access a
   URI and the access fails, the server SHOULD provide the failed URI in
   this header field in the method response, unless there are multiple
   URI failures, in which case the server MUST provide one of the failed
   URIs in this header field in the method response.

   failed-uri          =   "Failed-URI" ":" absoluteURI CRLF

8.4.13.  Failed-URI-Cause



   When a synthesizer method needs a synthesizer to fetch or access a
   URI and the access fails, the server MUST provide the URI-specific or
   protocol-specific response code for the URI in the Failed-URI header
   field in the method response through this header field.  The value
   encoding is UTF-8 (RFC 3629 [RFC3629]) to accommodate any access
   protocol -- some access protocols might have a response string
   instead of a numeric response code.

   failed-uri-cause    =   "Failed-URI-Cause" ":" 1*UTFCHAR CRLF








Burnett & Shanmugham         Standards Track                   [Page 55]

RFC 6787                         MRCPv2                    November 2012


8.4.14.  Speak-Restart



   When a client issues a CONTROL request to a currently speaking
   synthesizer resource to jump backward, and the target jump point is
   before the start of the current SPEAK request, the current SPEAK
   request MUST restart from the beginning of its speech data and the
   server's response to the CONTROL request MUST contain this header
   field with a value of "true" indicating a restart.

   speak-restart       =   "Speak-Restart" ":" BOOLEAN CRLF

8.4.15.  Speak-Length



   This header field MAY be specified in a CONTROL method to control the
   maximum length of speech to speak, relative to the current speaking
   point in the currently active SPEAK request.  If numeric, the value
   MUST be a positive integer.  If a header field with a Tag unit is
   specified, then the speech output continues until the tag is reached
   or the SPEAK request is completed, whichever comes first.  This
   header field MAY be specified in a SPEAK request to indicate the
   length to speak from the speech data and is relative to the point in
   speech that the SPEAK request starts.  The different speech length
   units supported are synthesizer implementation dependent.  If a
   server does not support the specified unit, the server MUST respond
   with a status-code of 409 "Unsupported Header Field Value".

   speak-length          =   "Speak-Length" ":" positive-length-value
                             CRLF

   positive-length-value =   positive-speech-length
                         /   text-speech-length

   text-speech-length    =   1*UTFCHAR SP "Tag"

   positive-speech-length =  1*19DIGIT SP numeric-speech-unit

   numeric-speech-unit   =   "Second"
                         /   "Word"
                         /   "Sentence"
                         /   "Paragraph"











Burnett & Shanmugham         Standards Track                   [Page 56]

RFC 6787                         MRCPv2                    November 2012


8.4.16.  Load-Lexicon



   This header field is used to indicate whether a lexicon has to be
   loaded or unloaded.  The value "true" means to load the lexicon if
   not already loaded, and the value "false" means to unload the lexicon
   if it is loaded.  The default value for this header field is "true".
   This header field MAY be specified in a DEFINE-LEXICON method.

   load-lexicon       =   "Load-Lexicon" ":" BOOLEAN CRLF

8.4.17.  Lexicon-Search-Order



   This header field is used to specify a list of active pronunciation
   lexicon URIs and the search order among the active lexicons.
   Lexicons specified within the SSML document take precedence over the
   lexicons specified in this header field.  This header field MAY be
   specified in the SPEAK, SET-PARAMS, and GET-PARAMS methods.

   lexicon-search-order =   "Lexicon-Search-Order" ":"
             "<" absoluteURI ">" *(" " "<" absoluteURI ">") CRLF

8.5.  Synthesizer Message Body



   A synthesizer message can contain additional information associated
   with the Request, Response, or Event in its message body.

8.5.1.  Synthesizer Speech Data



   Marked-up text for the synthesizer to speak is specified as a typed
   media entity in the message body.  The speech data to be spoken by
   the synthesizer can be specified inline by embedding the data in the
   message body or by reference by providing a URI for accessing the
   data.  In either case, the data and the format used to markup the
   speech needs to be of a content type supported by the server.

   All MRCPv2 servers containing synthesizer resources MUST support both
   plain text speech data and W3C's Speech Synthesis Markup Language
   [W3C.REC-speech-synthesis-20040907] and hence MUST support the media
   types 'text/plain' and 'application/ssml+xml'.  Other formats MAY be
   supported.

   If the speech data is to be fetched by URI reference, the media type
   'text/uri-list' (see RFC 2483 [RFC2483]) is used to indicate one or
   more URIs that, when dereferenced, will contain the content to be
   spoken.  If a list of speech URIs is specified, the resource MUST
   speak the speech data provided by each URI in the order in which the
   URIs are specified in the content.




Burnett & Shanmugham         Standards Track                   [Page 57]

RFC 6787                         MRCPv2                    November 2012


   MRCPv2 clients and servers MUST support the 'multipart/mixed' media
   type.  This is the appropriate media type to use when providing a mix
   of URI and inline speech data.  Embedded within the multipart content
   block, there MAY be content for the 'text/uri-list', 'application/
   ssml+xml', and/or 'text/plain' media types.  The character set and
   encoding used in the speech data is specified according to standard
   media type definitions.  The multipart content MAY also contain
   actual audio data.  Clients may have recorded audio clips stored in
   memory or on a local device and wish to play it as part of the SPEAK
   request.  The audio portions MAY be sent by the client as part of the
   multipart content block.  This audio is referenced in the speech
   markup data that is another part in the multipart content block
   according to the 'multipart/mixed' media type specification.

   Content-Type:text/uri-list
   Content-Length:...

   http://www.example.com/ASR-Introduction.ssml
   http://www.example.com/ASR-Document-Part1.ssml
   http://www.example.com/ASR-Document-Part2.ssml
   http://www.example.com/ASR-Conclusion.ssml

                             URI List Example


   Content-Type:application/ssml+xml
   Content-Length:...

   <?xml version="1.0"?>
        <speak version="1.0"
               xmlns="http://www.w3.org/2001/10/synthesis"
               xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
               xsi:schemaLocation="http://www.w3.org/2001/10/synthesis
                   http://www.w3.org/TR/speech-synthesis/synthesis.xsd"
               xml:lang="en-US">
          <p>
            <s>You have 4 new messages.</s>
            <s>The first is from Aldine Turnbet
            and arrived at <break/>
            <say-as interpret-as="vxml:time">0345p</say-as>.</s>

            <s>The subject is <prosody
            rate="-20%">ski trip</prosody></s>
         </p>
        </speak>

                               SSML Example




Burnett & Shanmugham         Standards Track                   [Page 58]

RFC 6787                         MRCPv2                    November 2012


   Content-Type:multipart/mixed; boundary="break"

   --break
   Content-Type:text/uri-list
   Content-Length:...

   http://www.example.com/ASR-Introduction.ssml
   http://www.example.com/ASR-Document-Part1.ssml
   http://www.example.com/ASR-Document-Part2.ssml
   http://www.example.com/ASR-Conclusion.ssml

   --break
   Content-Type:application/ssml+xml
   Content-Length:...

   <?xml version="1.0"?>
       <speak version="1.0"
              xmlns="http://www.w3.org/2001/10/synthesis"
              xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
              xsi:schemaLocation="http://www.w3.org/2001/10/synthesis
                   http://www.w3.org/TR/speech-synthesis/synthesis.xsd"
              xml:lang="en-US">
          <p>
            <s>You have 4 new messages.</s>
            <s>The first is from Stephanie Williams
            and arrived at <break/>
            <say-as interpret-as="vxml:time">0342p</say-as>.</s>

            <s>The subject is <prosody
            rate="-20%">ski trip</prosody></s>
          </p>
       </speak>
   --break--

                             Multipart Example

8.5.2.  Lexicon Data



   Synthesizer lexicon data from the client to the server can be
   provided inline or by reference.  Either way, they are carried as
   typed media in the message body of the MRCPv2 request message (see
   Section 8.14).

   When a lexicon is specified inline in the message, the client MUST
   provide a Content-ID for that lexicon as part of the content header
   fields.  The server MUST store the lexicon associated with that
   Content-ID for the duration of the session.  A stored lexicon can be
   overwritten by defining a new lexicon with the same Content-ID.



Burnett & Shanmugham         Standards Track                   [Page 59]

RFC 6787                         MRCPv2                    November 2012


   Lexicons that have been associated with a Content-ID can be
   referenced through the 'session' URI scheme (see Section 13.6).

   If lexicon data is specified by external URI reference, the media
   type 'text/uri-list' (see RFC 2483 [RFC2483] ) is used to list the
   one or more URIs that may be dereferenced to obtain the lexicon data.
   All MRCPv2 servers MUST support the "http" and "https" URI access
   mechanisms, and MAY support other mechanisms.

   If the data in the message body consists of a mix of URI and inline
   lexicon data, the 'multipart/mixed' media type is used.  The
   character set and encoding used in the lexicon data may be specified
   according to standard media type definitions.

8.6.  SPEAK Method



   The SPEAK request provides the synthesizer resource with the speech
   text and initiates speech synthesis and streaming.  The SPEAK method
   MAY carry voice and prosody header fields that alter the behavior of
   the voice being synthesized, as well as a typed media message body
   containing the actual marked-up text to be spoken.

   The SPEAK method implementation MUST do a fetch of all external URIs
   that are part of that operation.  If caching is implemented, this URI
   fetching MUST conform to the cache-control hints and parameter header
   fields associated with the method in deciding whether it is to be
   fetched from cache or from the external server.  If these hints/
   parameters are not specified in the method, the values set for the
   session using SET-PARAMS/GET-PARAMS apply.  If it was not set for the
   session, their default values apply.

   When applying voice parameters, there are three levels of precedence.
   The highest precedence are those specified within the speech markup
   text, followed by those specified in the header fields of the SPEAK
   request and hence that apply for that SPEAK request only, followed by
   the session default values that can be set using the SET-PARAMS
   request and apply for subsequent methods invoked during the session.

   If the resource was idle at the time the SPEAK request arrived at the
   server and the SPEAK method is being actively processed, the resource
   responds immediately with a success status code and a request-state
   of IN-PROGRESS.

   If the resource is in the speaking or paused state when the SPEAK
   method arrives at the server, i.e., it is in the middle of processing
   a previous SPEAK request, the status returns success with a request-
   state of PENDING.  The server places the SPEAK request in the
   synthesizer resource request queue.  The request queue operates



Burnett & Shanmugham         Standards Track                   [Page 60]

RFC 6787                         MRCPv2                    November 2012


   strictly FIFO: requests are processed serially in order of receipt.
   If the current SPEAK fails, all SPEAK methods in the pending queue
   are cancelled and each generates a SPEAK-COMPLETE event with a
   Completion-Cause of "cancelled".

   For the synthesizer resource, SPEAK is the only method that can
   return a request-state of IN-PROGRESS or PENDING.  When the text has
   been synthesized and played into the media stream, the resource
   issues a SPEAK-COMPLETE event with the request-id of the SPEAK
   request and a request-state of COMPLETE.

   C->S: MRCP/2.0 ... SPEAK 543257
         Channel-Identifier:32AECB23433802@speechsynth
         Voice-gender:neutral
         Voice-Age:25
         Prosody-volume:medium
         Content-Type:application/ssml+xml
         Content-Length:...

         <?xml version="1.0"?>
            <speak version="1.0"
                xmlns="http://www.w3.org/2001/10/synthesis"
                xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
                xsi:schemaLocation="http://www.w3.org/2001/10/synthesis
                   http://www.w3.org/TR/speech-synthesis/synthesis.xsd"
                xml:lang="en-US">
            <p>
             <s>You have 4 new messages.</s>
             <s>The first is from Stephanie Williams and arrived at
                <break/>
                <say-as interpret-as="vxml:time">0342p</say-as>.
                </s>
             <s>The subject is
                    <prosody rate="-20%">ski trip</prosody>
             </s>
            </p>
           </speak>

   S->C: MRCP/2.0 ... 543257 200 IN-PROGRESS
         Channel-Identifier:32AECB23433802@speechsynth
         Speech-Marker:timestamp=857206027059

   S->C: MRCP/2.0 ... SPEAK-COMPLETE 543257 COMPLETE
         Channel-Identifier:32AECB23433802@speechsynth
         Completion-Cause:000 normal
         Speech-Marker:timestamp=857206027059

                               SPEAK Example



Burnett & Shanmugham         Standards Track                   [Page 61]

RFC 6787                         MRCPv2                    November 2012


8.7.  STOP



   The STOP method from the client to the server tells the synthesizer
   resource to stop speaking if it is speaking something.

   The STOP request can be sent with an Active-Request-Id-List header
   field to stop the zero or more specific SPEAK requests that may be in
   queue and return a response status-code of 200 "Success".  If no
   Active-Request-Id-List header field is sent in the STOP request, the
   server terminates all outstanding SPEAK requests.

   If a STOP request successfully terminated one or more PENDING or
   IN-PROGRESS SPEAK requests, then the response MUST contain an Active-
   Request-Id-List header field enumerating the SPEAK request-ids that
   were terminated.  Otherwise, there is no Active-Request-Id-List
   header field in the response.  No SPEAK-COMPLETE events are sent for
   such terminated requests.

   If a SPEAK request that was IN-PROGRESS and speaking was stopped, the
   next pending SPEAK request, if any, becomes IN-PROGRESS at the
   resource and enters the speaking state.

   If a SPEAK request that was IN-PROGRESS and paused was stopped, the
   next pending SPEAK request, if any, becomes IN-PROGRESS and enters
   the paused state.

   C->S: MRCP/2.0 ... SPEAK 543258
         Channel-Identifier:32AECB23433802@speechsynth
         Content-Type:application/ssml+xml
         Content-Length:...

         <?xml version="1.0"?>
           <speak version="1.0"
                xmlns="http://www.w3.org/2001/10/synthesis"
                xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
                xsi:schemaLocation="http://www.w3.org/2001/10/synthesis
                   http://www.w3.org/TR/speech-synthesis/synthesis.xsd"
                xml:lang="en-US">
            <p>
             <s>You have 4 new messages.</s>
             <s>The first is from Stephanie Williams and arrived at
                <break/>
                <say-as interpret-as="vxml:time">0342p</say-as>.</s>
             <s>The subject is
                 <prosody rate="-20%">ski trip</prosody></s>
            </p>
           </speak>




Burnett & Shanmugham         Standards Track                   [Page 62]

RFC 6787                         MRCPv2                    November 2012


   S->C: MRCP/2.0 ... 543258 200 IN-PROGRESS
         Channel-Identifier:32AECB23433802@speechsynth
         Speech-Marker:timestamp=857206027059

   C->S: MRCP/2.0 ... STOP 543259
         Channel-Identifier:32AECB23433802@speechsynth

   S->C: MRCP/2.0 ... 543259 200 COMPLETE
         Channel-Identifier:32AECB23433802@speechsynth
         Active-Request-Id-List:543258
         Speech-Marker:timestamp=857206039059

                               STOP Example

8.8.  BARGE-IN-OCCURRED



   The BARGE-IN-OCCURRED method, when used with the synthesizer
   resource, provides a client that has detected a barge-in-able event a
   means to communicate the occurrence of the event to the synthesizer
   resource.

   This method is useful in two scenarios:

   1.  The client has detected DTMF digits in the input media or some
       other barge-in-able event and wants to communicate that to the
       synthesizer resource.

   2.  The recognizer resource and the synthesizer resource are in
       different servers.  In this case, the client acts as an
       intermediary for the two servers.  It receives an event from the
       recognition resource and sends a BARGE-IN-OCCURRED request to the
       synthesizer.  In such cases, the BARGE-IN-OCCURRED method would
       also have a Proxy-Sync-Id header field received from the resource
       generating the original event.

   If a SPEAK request is active with kill-on-barge-in enabled (see
   Section 8.4.2), and the BARGE-IN-OCCURRED event is received, the
   synthesizer MUST immediately stop streaming out audio.  It MUST also
   terminate any speech requests queued behind the current active one,
   irrespective of whether or not they have barge-in enabled.  If a
   barge-in-able SPEAK request was playing and it was terminated, the
   response MUST contain an Active-Request-Id-List header field listing
   the request-ids of all SPEAK requests that were terminated.  The
   server generates no SPEAK-COMPLETE events for these requests.







Burnett & Shanmugham         Standards Track                   [Page 63]

RFC 6787                         MRCPv2                    November 2012


   If there were no SPEAK requests terminated by the synthesizer
   resource as a result of the BARGE-IN-OCCURRED method, the server MUST
   respond to the BARGE-IN-OCCURRED with a status-code of 200 "Success",
   and the response MUST NOT contain an Active-Request-Id-List header
   field.

   If the synthesizer and recognizer resources are part of the same
   MRCPv2 session, they can be optimized for a quicker kill-on-barge-in
   response if the recognizer and synthesizer interact directly.  In
   these cases, the client MUST still react to a START-OF-INPUT event
   from the recognizer by invoking the BARGE-IN-OCCURRED method to the
   synthesizer.  The client MUST invoke the BARGE-IN-OCCURRED if it has
   any outstanding requests to the synthesizer resource in either the
   PENDING or IN-PROGRESS state.

   C->S: MRCP/2.0 ... SPEAK 543258
         Channel-Identifier:32AECB23433802@speechsynth
         Voice-gender:neutral
         Voice-Age:25
         Prosody-volume:medium
         Content-Type:application/ssml+xml
         Content-Length:...

         <?xml version="1.0"?>
           <speak version="1.0"
                xmlns="http://www.w3.org/2001/10/synthesis"
                xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
                xsi:schemaLocation="http://www.w3.org/2001/10/synthesis
                   http://www.w3.org/TR/speech-synthesis/synthesis.xsd"
                xml:lang="en-US">
            <p>
             <s>You have 4 new messages.</s>
             <s>The first is from Stephanie Williams and arrived at
                <break/>
                <say-as interpret-as="vxml:time">0342p</say-as>.</s>
             <s>The subject is
                <prosody rate="-20%">ski trip</prosody></s>
            </p>
           </speak>

   S->C: MRCP/2.0 ... 543258 200 IN-PROGRESS
         Channel-Identifier:32AECB23433802@speechsynth
         Speech-Marker:timestamp=857206027059

   C->S: MRCP/2.0 ... BARGE-IN-OCCURRED 543259
         Channel-Identifier:32AECB23433802@speechsynth
         Proxy-Sync-Id:987654321




Burnett & Shanmugham         Standards Track                   [Page 64]

RFC 6787                         MRCPv2                    November 2012


   S->C:MRCP/2.0 ... 543259 200 COMPLETE
         Channel-Identifier:32AECB23433802@speechsynth
         Active-Request-Id-List:543258
         Speech-Marker:timestamp=857206039059

                         BARGE-IN-OCCURRED Example

8.9.  PAUSE



   The PAUSE method from the client to the server tells the synthesizer
   resource to pause speech output if it is speaking something.  If a
   PAUSE method is issued on a session when a SPEAK is not active, the
   server MUST respond with a status-code of 402 "Method not valid in
   this state".  If a PAUSE method is issued on a session when a SPEAK
   is active and paused, the server MUST respond with a status-code of
   200 "Success".  If a SPEAK request was active, the server MUST return
   an Active-Request-Id-List header field whose value contains the
   request-id of the SPEAK request that was paused.

   C->S: MRCP/2.0 ... SPEAK 543258
         Channel-Identifier:32AECB23433802@speechsynth
         Voice-gender:neutral
         Voice-Age:25
         Prosody-volume:medium
         Content-Type:application/ssml+xml
         Content-Length:...

         <?xml version="1.0"?>
           <speak version="1.0"
                xmlns="http://www.w3.org/2001/10/synthesis"
                xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
                xsi:schemaLocation="http://www.w3.org/2001/10/synthesis
                   http://www.w3.org/TR/speech-synthesis/synthesis.xsd"
                xml:lang="en-US">
            <p>
             <s>You have 4 new messages.</s>
             <s>The first is from Stephanie Williams and arrived at
                <break/>
                <say-as interpret-as="vxml:time">0342p</say-as>.</s>

             <s>The subject is
                <prosody rate="-20%">ski trip</prosody></s>
            </p>
           </speak>

   S->C: MRCP/2.0 ... 543258 200 IN-PROGRESS
         Channel-Identifier:32AECB23433802@speechsynth
         Speech-Marker:timestamp=857206027059



Burnett & Shanmugham         Standards Track                   [Page 65]

RFC 6787                         MRCPv2                    November 2012


   C->S: MRCP/2.0 ... PAUSE 543259
         Channel-Identifier:32AECB23433802@speechsynth

   S->C: MRCP/2.0 ... 543259 200 COMPLETE
         Channel-Identifier:32AECB23433802@speechsynth
         Active-Request-Id-List:543258

                               PAUSE Example

8.10.  RESUME



   The RESUME method from the client to the server tells a paused
   synthesizer resource to resume speaking.  If a RESUME request is
   issued on a session with no active SPEAK request, the server MUST
   respond with a status-code of 402 "Method not valid in this state".
   If a RESUME request is issued on a session with an active SPEAK
   request that is speaking (i.e., not paused), the server MUST respond
   with a status-code of 200 "Success".  If a SPEAK request was paused,
   the server MUST return an Active-Request-Id-List header field whose
   value contains the request-id of the SPEAK request that was resumed.

   C->S: MRCP/2.0 ... SPEAK 543258
         Channel-Identifier:32AECB23433802@speechsynth
         Voice-gender:neutral
         Voice-age:25
         Prosody-volume:medium
         Content-Type:application/ssml+xml
         Content-Length:...

         <?xml version="1.0"?>
           <speak version="1.0"
                xmlns="http://www.w3.org/2001/10/synthesis"
                xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
                xsi:schemaLocation="http://www.w3.org/2001/10/synthesis
                   http://www.w3.org/TR/speech-synthesis/synthesis.xsd"
                xml:lang="en-US">
            <p>
             <s>You have 4 new messages.</s>
             <s>The first is from Stephanie Williams and arrived at
                <break/>
                <say-as interpret-as="vxml:time">0342p</say-as>.</s>
             <s>The subject is
                <prosody rate="-20%">ski trip</prosody></s>
            </p>
           </speak>






Burnett & Shanmugham         Standards Track                   [Page 66]

RFC 6787                         MRCPv2                    November 2012


   S->C: MRCP/2.0 ... 543258 200 IN-PROGRESS@speechsynth
         Channel-Identifier:32AECB23433802
         Speech-Marker:timestamp=857206027059

   C->S: MRCP/2.0 ... PAUSE 543259
         Channel-Identifier:32AECB23433802@speechsynth

   S->C: MRCP/2.0 ... 543259 200 COMPLETE
         Channel-Identifier:32AECB23433802@speechsynth
         Active-Request-Id-List:543258

   C->S: MRCP/2.0 ... RESUME 543260
         Channel-Identifier:32AECB23433802@speechsynth

   S->C: MRCP/2.0 ... 543260 200 COMPLETE
         Channel-Identifier:32AECB23433802@speechsynth
         Active-Request-Id-List:543258

                              RESUME Example

8.11.  CONTROL



   The CONTROL method from the client to the server tells a synthesizer
   that is speaking to modify what it is speaking on the fly.  This
   method is used to request the synthesizer to jump forward or backward
   in what it is speaking, change speaker rate, speaker parameters, etc.
   It affects only the currently IN-PROGRESS SPEAK request.  Depending
   on the implementation and capability of the synthesizer resource, it
   may or may not support the various modifications indicated by header
   fields in the CONTROL request.

   When a client invokes a CONTROL method to jump forward and the
   operation goes beyond the end of the active SPEAK method's text, the
   CONTROL request still succeeds.  The active SPEAK request completes
   and returns a SPEAK-COMPLETE event following the response to the
   CONTROL method.  If there are more SPEAK requests in the queue, the
   synthesizer resource starts at the beginning of the next SPEAK
   request in the queue.

   When a client invokes a CONTROL method to jump backward and the
   operation jumps to the beginning or beyond the beginning of the
   speech data of the active SPEAK method, the CONTROL request still
   succeeds.  The response to the CONTROL request contains the speak-
   restart header field, and the active SPEAK request restarts from the
   beginning of its speech data.






Burnett & Shanmugham         Standards Track                   [Page 67]

RFC 6787                         MRCPv2                    November 2012


   These two behaviors can be used to rewind or fast-forward across
   multiple speech requests, if the client wants to break up a speech
   markup text into multiple SPEAK requests.

   If a SPEAK request was active when the CONTROL method was received,
   the server MUST return an Active-Request-Id-List header field
   containing the request-id of the SPEAK request that was active.

   C->S: MRCP/2.0 ... SPEAK 543258
         Channel-Identifier:32AECB23433802@speechsynth
         Voice-gender:neutral
         Voice-age:25
         Prosody-volume:medium
         Content-Type:application/ssml+xml
         Content-Length:...

         <?xml version="1.0"?>
           <speak version="1.0"
                xmlns="http://www.w3.org/2001/10/synthesis"
                xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
                xsi:schemaLocation="http://www.w3.org/2001/10/synthesis
                   http://www.w3.org/TR/speech-synthesis/synthesis.xsd"
                xml:lang="en-US">
            <p>
             <s>You have 4 new messages.</s>
             <s>The first is from Stephanie Williams
                and arrived at <break/>
                <say-as interpret-as="vxml:time">0342p</say-as>.</s>

             <s>The subject is <prosody
                rate="-20%">ski trip</prosody></s>
            </p>
           </speak>

   S->C: MRCP/2.0 ... 543258 200 IN-PROGRESS
         Channel-Identifier:32AECB23433802@speechsynth
         Speech-Marker:timestamp=857205016059

   C->S: MRCP/2.0 ... CONTROL 543259
         Channel-Identifier:32AECB23433802@speechsynth
         Prosody-rate:fast

   S->C: MRCP/2.0 ... 543259 200 COMPLETE
         Channel-Identifier:32AECB23433802@speechsynth
         Active-Request-Id-List:543258
         Speech-Marker:timestamp=857206027059





Burnett & Shanmugham         Standards Track                   [Page 68]

RFC 6787                         MRCPv2                    November 2012


   C->S: MRCP/2.0 ... CONTROL 543260
         Channel-Identifier:32AECB23433802@speechsynth
         Jump-Size:-15 Words

   S->C: MRCP/2.0 ... 543260 200 COMPLETE
         Channel-Identifier:32AECB23433802@speechsynth
         Active-Request-Id-List:543258
         Speech-Marker:timestamp=857206039059

                              CONTROL Example

8.12.  SPEAK-COMPLETE



   This is an Event message from the synthesizer resource to the client
   that indicates the corresponding SPEAK request was completed.  The
   request-id field matches the request-id of the SPEAK request that
   initiated the speech that just completed.  The request-state field is
   set to COMPLETE by the server, indicating that this is the last event
   with the corresponding request-id.  The Completion-Cause header field
   specifies the cause code pertaining to the status and reason of
   request completion, such as the SPEAK completed normally or because
   of an error, kill-on-barge-in, etc.

   C->S: MRCP/2.0 ... SPEAK 543260
         Channel-Identifier:32AECB23433802@speechsynth
         Voice-gender:neutral
         Voice-age:25
         Prosody-volume:medium
         Content-Type:application/ssml+xml
         Content-Length:...

         <?xml version="1.0"?>
           <speak version="1.0"
                xmlns="http://www.w3.org/2001/10/synthesis"
                xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
                xsi:schemaLocation="http://www.w3.org/2001/10/synthesis
                   http://www.w3.org/TR/speech-synthesis/synthesis.xsd"
                xml:lang="en-US">
            <p>
             <s>You have 4 new messages.</s>
             <s>The first is from Stephanie Williams
                and arrived at <break/>
                <say-as interpret-as="vxml:time">0342p</say-as>.</s>
             <s>The subject is
                <prosody rate="-20%">ski trip</prosody></s>
            </p>
           </speak>




Burnett & Shanmugham         Standards Track                   [Page 69]

RFC 6787                         MRCPv2                    November 2012


   S->C: MRCP/2.0 ... 543260 200 IN-PROGRESS
         Channel-Identifier:32AECB23433802@speechsynth
         Speech-Marker:timestamp=857206027059

   S->C: MRCP/2.0 ... SPEAK-COMPLETE 543260 COMPLETE
         Channel-Identifier:32AECB23433802@speechsynth
         Completion-Cause:000 normal
         Speech-Marker:timestamp=857206039059

                          SPEAK-COMPLETE Example

8.13.  SPEECH-MARKER



   This is an event generated by the synthesizer resource to the client
   when the synthesizer encounters a marker tag in the speech markup it
   is currently processing.  The value of the request-id field MUST
   match that of the corresponding SPEAK request.  The request-state
   field MUST have the value "IN-PROGRESS" as the speech is still not
   complete.  The value of the speech marker tag hit, describing where
   the synthesizer is in the speech markup, MUST be returned in the
   Speech-Marker header field, along with an NTP timestamp indicating
   the instant in the output speech stream that the marker was
   encountered.  The SPEECH-MARKER event MUST also be generated with a
   null marker value and output NTP timestamp when a SPEAK request in
   Pending-State (i.e., in the queue) changes state to IN-PROGRESS and
   starts speaking.  The NTP timestamp MUST be synchronized with the RTP
   timestamp used to generate the speech stream through standard RTCP
   machinery.

   C->S: MRCP/2.0 ... SPEAK 543261
         Channel-Identifier:32AECB23433802@speechsynth
         Voice-gender:neutral
         Voice-age:25
         Prosody-volume:medium
         Content-Type:application/ssml+xml
         Content-Length:...

         <?xml version="1.0"?>
           <speak version="1.0"
                xmlns="http://www.w3.org/2001/10/synthesis"
                xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
                xsi:schemaLocation="http://www.w3.org/2001/10/synthesis
                   http://www.w3.org/TR/speech-synthesis/synthesis.xsd"
                xml:lang="en-US">
            <p>
             <s>You have 4 new messages.</s>
             <s>The first is from Stephanie Williams
                and arrived at <break/>



Burnett & Shanmugham         Standards Track                   [Page 70]

RFC 6787                         MRCPv2                    November 2012


                <say-as interpret-as="vxml:time">0342p</say-as>.</s>
                <mark name="here"/>
             <s>The subject is
                <prosody rate="-20%">ski trip</prosody>
             </s>
             <mark name="ANSWER"/>
            </p>
           </speak>

   S->C: MRCP/2.0 ... 543261 200 IN-PROGRESS
         Channel-Identifier:32AECB23433802@speechsynth
         Speech-Marker:timestamp=857205015059

   S->C: MRCP/2.0 ... SPEECH-MARKER 543261 IN-PROGRESS
         Channel-Identifier:32AECB23433802@speechsynth
         Speech-Marker:timestamp=857206027059;here

   S->C: MRCP/2.0 ... SPEECH-MARKER 543261 IN-PROGRESS
         Channel-Identifier:32AECB23433802@speechsynth
         Speech-Marker:timestamp=857206039059;ANSWER

   S->C: MRCP/2.0 ... SPEAK-COMPLETE 543261 COMPLETE
         Channel-Identifier:32AECB23433802@speechsynth
         Completion-Cause:000 normal
         Speech-Marker:timestamp=857207689259;ANSWER

                           SPEECH-MARKER Example

8.14.  DEFINE-LEXICON



   The DEFINE-LEXICON method, from the client to the server, provides a
   lexicon and tells the server to load or unload the lexicon (see
   Section 8.4.16).  The media type of the lexicon is provided in the
   Content-Type header (see Section 8.5.2).  One such media type is
   "application/pls+xml" for the Pronunciation Lexicon Specification
   (PLS) [W3C.REC-pronunciation-lexicon-20081014] [RFC4267].

   If the server resource is in the speaking or paused state, the server
   MUST respond with a failure status-code of 402 "Method not valid in
   this state".

   If the resource is in the idle state and is able to successfully
   load/unload the lexicon, the status MUST return a 200 "Success"
   status-code and the request-state MUST be COMPLETE.







Burnett & Shanmugham         Standards Track                   [Page 71]

RFC 6787                         MRCPv2                    November 2012


   If the synthesizer could not define the lexicon for some reason, for
   example, because the download failed or the lexicon was in an
   unsupported form, the server MUST respond with a failure status-code
   of 407 and a Completion-Cause header field describing the failure
   reason.

9.  Speech Recognizer Resource



   The speech recognizer resource receives an incoming voice stream and
   provides the client with an interpretation of what was spoken in
   textual form.

   The recognizer resource is controlled by MRCPv2 requests from the
   client.  The recognizer resource can both respond to these requests
   and generate asynchronous events to the client to indicate conditions
   of interest during the processing of the method.

   This section applies to the following resource types.

   1.  speechrecog

   2.  dtmfrecog

   The difference between the above two resources is in their level of
   support for recognition grammars.  The "dtmfrecog" resource type is
   capable of recognizing only DTMF digits and hence accepts only DTMF
   grammars.  It only generates barge-in for DTMF inputs and ignores
   speech.  The "speechrecog" resource type can recognize regular speech
   as well as DTMF digits and hence MUST support grammars describing
   either speech or DTMF.  This resource generates barge-in events for
   speech and/or DTMF.  By analyzing the grammars that are activated by
   the RECOGNIZE method, it determines if a barge-in should occur for
   speech and/or DTMF.  When the recognizer decides it needs to generate
   a barge-in, it also generates a START-OF-INPUT event to the client.
   The recognizer resource MAY support recognition in the normal or
   hotword modes or both (although note that a single "speechrecog"
   resource does not perform normal and hotword mode recognition
   simultaneously).  For implementations where a single recognizer
   resource does not support both modes, or simultaneous normal and
   hotword recognition is desired, the two modes can be invoked through
   separate resources allocated to the same SIP dialog (with different
   MRCP session identifiers) and share the RTP audio feed.

   The capabilities of the recognizer resource are enumerated below:

   Normal Mode Recognition  Normal mode recognition tries to match all
      of the speech or DTMF against the grammar and returns a no-match
      status if the input fails to match or the method times out.



Burnett & Shanmugham         Standards Track                   [Page 72]

RFC 6787                         MRCPv2                    November 2012


   Hotword Mode Recognition  Hotword mode is where the recognizer looks
      for a match against specific speech grammar or DTMF sequence and
      ignores speech or DTMF that does not match.  The recognition
      completes only if there is a successful match of grammar, if the
      client cancels the request, or if there is a non-input or
      recognition timeout.

   Voice Enrolled Grammars  A recognizer resource MAY optionally support
      Voice Enrolled Grammars.  With this functionality, enrollment is
      performed using a person's voice.  For example, a list of contacts
      can be created and maintained by recording the person's names
      using the caller's voice.  This technique is sometimes also called
      speaker-dependent recognition.

   Interpretation  A recognizer resource MAY be employed strictly for
      its natural language interpretation capabilities by supplying it
      with a text string as input instead of speech.  In this mode, the
      resource takes text as input and produces an "interpretation" of
      the input according to the supplied grammar.

   Voice enrollment has the concept of an enrollment session.  A session
   to add a new phrase to a personal grammar involves the initial
   enrollment followed by a repeat of enough utterances before
   committing the new phrase to the personal grammar.  Each time an
   utterance is recorded, it is compared for similarity with the other
   samples and a clash test is performed against other entries in the
   personal grammar to ensure there are no similar and confusable
   entries.

   Enrollment is done using a recognizer resource.  Controlling which
   utterances are to be considered for enrollment of a new phrase is
   done by setting a header field (see Section 9.4.39) in the Recognize
   request.

   Interpretation is accomplished through the INTERPRET method
   (Section 9.20) and the Interpret-Text header field (Section 9.4.30).















Burnett & Shanmugham         Standards Track                   [Page 73]

RFC 6787                         MRCPv2                    November 2012


9.1.  Recognizer State Machine



   The recognizer resource maintains a state machine to process MRCPv2
   requests from the client.

   Idle                   Recognizing               Recognized
   State                  State                     State
    |                       |                          |
    |---------RECOGNIZE---->|---RECOGNITION-COMPLETE-->|
    |<------STOP------------|<-----RECOGNIZE-----------|
    |                       |                          |
    |              |--------|              |-----------|
    |       START-OF-INPUT  |       GET-RESULT         |
    |              |------->|              |---------->|
    |------------|          |                          |
    |      DEFINE-GRAMMAR   |----------|               |
    |<-----------|          | START-INPUT-TIMERS       |
    |                       |<---------|               |
    |------|                |                          |
    |  INTERPRET            |                          |
    |<-----|                |------|                   |
    |                       |   RECOGNIZE              |
    |-------|               |<-----|                   |
    |      STOP                                        |
    |<------|                                          |
    |<-------------------STOP--------------------------|
    |<-------------------DEFINE-GRAMMAR----------------|

                         Recognizer State Machine

   If a recognizer resource supports voice enrolled grammars, starting
   an enrollment session does not change the state of the recognizer
   resource.  Once an enrollment session is started, then utterances are
   enrolled by calling the RECOGNIZE method repeatedly.  The state of
   the speech recognizer resource goes from IDLE to RECOGNIZING state
   each time RECOGNIZE is called.

9.2.  Recognizer Methods



   The recognizer supports the following methods.

   recognizer-method    =  recog-only-method
                        /  enrollment-method








Burnett & Shanmugham         Standards Track                   [Page 74]

RFC 6787                         MRCPv2                    November 2012


   recog-only-method    =  "DEFINE-GRAMMAR"
                        /  "RECOGNIZE"
                        /  "INTERPRET"
                        /  "GET-RESULT"
                        /  "START-INPUT-TIMERS"
                        /  "STOP"

   It is OPTIONAL for a recognizer resource to support voice enrolled
   grammars.  If the recognizer resource does support voice enrolled
   grammars, it MUST support the following methods.

   enrollment-method    =  "START-PHRASE-ENROLLMENT"
                        /  "ENROLLMENT-ROLLBACK"
                        /  "END-PHRASE-ENROLLMENT"
                        /  "MODIFY-PHRASE"
                        /  "DELETE-PHRASE"

9.3.  Recognizer Events



   The recognizer can generate the following events.

   recognizer-event     =  "START-OF-INPUT"
                        /  "RECOGNITION-COMPLETE"
                        /  "INTERPRETATION-COMPLETE"

9.4.  Recognizer Header Fields



   A recognizer message can contain header fields containing request
   options and information to augment the Method, Response, or Event
   message it is associated with.

   recognizer-header    =  recog-only-header
                        /  enrollment-header

   recog-only-header    =  confidence-threshold
                        /  sensitivity-level
                        /  speed-vs-accuracy
                        /  n-best-list-length
                        /  no-input-timeout
                        /  input-type
                        /  recognition-timeout
                        /  waveform-uri
                        /  input-waveform-uri
                        /  completion-cause
                        /  completion-reason
                        /  recognizer-context-block
                        /  start-input-timers
                        /  speech-complete-timeout



Burnett & Shanmugham         Standards Track                   [Page 75]

RFC 6787                         MRCPv2                    November 2012


                        /  speech-incomplete-timeout
                        /  dtmf-interdigit-timeout
                        /  dtmf-term-timeout
                        /  dtmf-term-char
                        /  failed-uri
                        /  failed-uri-cause
                        /  save-waveform
                        /  media-type
                        /  new-audio-channel
                        /  speech-language
                        /  ver-buffer-utterance
                        /  recognition-mode
                        /  cancel-if-queue
                        /  hotword-max-duration
                        /  hotword-min-duration
                        /  interpret-text
                        /  dtmf-buffer-time
                        /  clear-dtmf-buffer
                        /  early-no-match

   If a recognizer resource supports voice enrolled grammars, the
   following header fields are also used.

   enrollment-header    =  num-min-consistent-pronunciations
                        /  consistency-threshold
                        /  clash-threshold
                        /  personal-grammar-uri
                        /  enroll-utterance
                        /  phrase-id
                        /  phrase-nl
                        /  weight
                        /  save-best-waveform
                        /  new-phrase-id
                        /  confusable-phrases-uri
                        /  abort-phrase-enrollment

   For enrollment-specific header fields that can appear as part of
   SET-PARAMS or GET-PARAMS methods, the following general rule applies:
   the START-PHRASE-ENROLLMENT method MUST be invoked before these
   header fields may be set through the SET-PARAMS method or retrieved
   through the GET-PARAMS method.

   Note that the Waveform-URI header field of the Recognizer resource
   can also appear in the response to the END-PHRASE-ENROLLMENT method.







Burnett & Shanmugham         Standards Track                   [Page 76]

RFC 6787                         MRCPv2                    November 2012


9.4.1.  Confidence-Threshold



   When a recognizer resource recognizes or matches a spoken phrase with
   some portion of the grammar, it associates a confidence level with
   that match.  The Confidence-Threshold header field tells the
   recognizer resource what confidence level the client considers a
   successful match.  This is a float value between 0.0-1.0 indicating
   the recognizer's confidence in the recognition.  If the recognizer
   determines that there is no candidate match with a confidence that is
   greater than the confidence threshold, then it MUST return no-match
   as the recognition result.  This header field MAY occur in RECOGNIZE,
   SET-PARAMS, or GET-PARAMS.  The default value for this header field
   is implementation specific, as is the interpretation of any specific
   value for this header field.  Although values for servers from
   different vendors are not comparable, it is expected that clients
   will tune this value over time for a given server.

   confidence-threshold     =  "Confidence-Threshold" ":" FLOAT CRLF

9.4.2.  Sensitivity-Level



   To filter out background noise and not mistake it for speech, the
   recognizer resource supports a variable level of sound sensitivity.
   The Sensitivity-Level header field is a float value between 0.0 and
   1.0 and allows the client to set the sensitivity level for the
   recognizer.  This header field MAY occur in RECOGNIZE, SET-PARAMS, or
   GET-PARAMS.  A higher value for this header field means higher
   sensitivity.  The default value for this header field is
   implementation specific, as is the interpretation of any specific
   value for this header field.  Although values for servers from
   different vendors are not comparable, it is expected that clients
   will tune this value over time for a given server.

   sensitivity-level        =  "Sensitivity-Level" ":" FLOAT CRLF

9.4.3.  Speed-Vs-Accuracy



   Depending on the implementation and capability of the recognizer
   resource it may be tunable towards Performance or Accuracy.  Higher
   accuracy may mean more processing and higher CPU utilization, meaning
   fewer active sessions per server and vice versa.  The value is a
   float between 0.0 and 1.0.  A value of 0.0 means fastest recognition.
   A value of 1.0 means best accuracy.  This header field MAY occur in
   RECOGNIZE, SET-PARAMS, or GET-PARAMS.  The default value for this







Burnett & Shanmugham         Standards Track                   [Page 77]

RFC 6787                         MRCPv2                    November 2012


   header field is implementation specific.  Although values for servers
   from different vendors are not comparable, it is expected that
   clients will tune this value over time for a given server.

   speed-vs-accuracy        =  "Speed-Vs-Accuracy" ":" FLOAT CRLF

9.4.4.  N-Best-List-Length



   When the recognizer matches an incoming stream with the grammar, it
   may come up with more than one alternative match because of
   confidence levels in certain words or conversation paths.  If this
   header field is not specified, by default, the recognizer resource
   returns only the best match above the confidence threshold.  The
   client, by setting this header field, can ask the recognition
   resource to send it more than one alternative.  All alternatives must
   still be above the Confidence-Threshold.  A value greater than one
   does not guarantee that the recognizer will provide the requested
   number of alternatives.  This header field MAY occur in RECOGNIZE,
   SET-PARAMS, or GET-PARAMS.  The minimum value for this header field
   is 1.  The default value for this header field is 1.

   n-best-list-length       =  "N-Best-List-Length" ":" 1*19DIGIT CRLF

9.4.5.  Input-Type



   When the recognizer detects barge-in-able input and generates a
   START-OF-INPUT event, that event MUST carry this header field to
   specify whether the input that caused the barge-in was DTMF or
   speech.

   input-type         =  "Input-Type" ":"  inputs CRLF
   inputs             =  "speech" / "dtmf"

9.4.6.  No-Input-Timeout



   When recognition is started and there is no speech detected for a
   certain period of time, the recognizer can send a RECOGNITION-
   COMPLETE event to the client with a Completion-Cause of "no-input-
   timeout" and terminate the recognition operation.  The client can use
   the No-Input-Timeout header field to set this timeout.  The value is
   in milliseconds and can range from 0 to an implementation-specific
   maximum value.  This header field MAY occur in RECOGNIZE, SET-PARAMS,
   or GET-PARAMS.  The default value is implementation specific.

   no-input-timeout         =  "No-Input-Timeout" ":" 1*19DIGIT CRLF






Burnett & Shanmugham         Standards Track                   [Page 78]

RFC 6787                         MRCPv2                    November 2012


9.4.7.  Recognition-Timeout



   When recognition is started and there is no match for a certain
   period of time, the recognizer can send a RECOGNITION-COMPLETE event
   to the client and terminate the recognition operation.  The
   Recognition-Timeout header field allows the client to set this
   timeout value.  The value is in milliseconds.  The value for this
   header field ranges from 0 to an implementation-specific maximum
   value.  The default value is 10 seconds.  This header field MAY occur
   in RECOGNIZE, SET-PARAMS, or GET-PARAMS.

   recognition-timeout      =  "Recognition-Timeout" ":" 1*19DIGIT CRLF

9.4.8.  Waveform-URI



   If the Save-Waveform header field is set to "true", the recognizer
   MUST record the incoming audio stream of the recognition into a
   stored form and provide a URI for the client to access it.  This
   header field MUST be present in the RECOGNITION-COMPLETE event if the
   Save-Waveform header field was set to "true".  The value of the
   header field MUST be empty if there was some error condition
   preventing the server from recording.  Otherwise, the URI generated
   by the server MUST be unambiguous across the server and all its
   recognition sessions.  The content associated with the URI MUST be
   available to the client until the MRCPv2 session terminates.

   Similarly, if the Save-Best-Waveform header field is set to "true",
   the recognizer MUST save the audio stream for the best repetition of
   the phrase that was used during the enrollment session.  The
   recognizer MUST then record the recognized audio and make it
   available to the client by returning a URI in the Waveform-URI header
   field in the response to the END-PHRASE-ENROLLMENT method.  The value
   of the header field MUST be empty if there was some error condition
   preventing the server from recording.  Otherwise, the URI generated
   by the server MUST be unambiguous across the server and all its
   recognition sessions.  The content associated with the URI MUST be
   available to the client until the MRCPv2 session terminates.  See the
   discussion on the sensitivity of saved waveforms in Section 12.

   The server MUST also return the size in octets and the duration in
   milliseconds of the recorded audio waveform as parameters associated
   with the header field.

   waveform-uri             =  "Waveform-URI" ":" ["<" uri ">"
                               ";" "size" "=" 1*19DIGIT
                               ";" "duration" "=" 1*19DIGIT] CRLF





Burnett & Shanmugham         Standards Track                   [Page 79]

RFC 6787                         MRCPv2                    November 2012


9.4.9.  Media-Type



   This header field MAY be specified in the SET-PARAMS, GET-PARAMS, or
   the RECOGNIZE methods and tells the server resource the media type in
   which to store captured audio or video, such as the one captured and
   returned by the Waveform-URI header field.

   media-type               =  "Media-Type" ":" media-type-value
                               CRLF

9.4.10.  Input-Waveform-URI



   This optional header field specifies a URI pointing to audio content
   to be processed by the RECOGNIZE operation.  This enables the client
   to request recognition from a specified buffer or audio file.

   input-waveform-uri       =  "Input-Waveform-URI" ":" uri CRLF

9.4.11.  Completion-Cause



   This header field MUST be part of a RECOGNITION-COMPLETE event coming
   from the recognizer resource to the client.  It indicates the reason
   behind the RECOGNIZE method completion.  This header field MUST be
   sent in the DEFINE-GRAMMAR and RECOGNIZE responses, if they return
   with a failure status and a COMPLETE state.  In the ABNF below, the
   cause-code contains a numerical value selected from the Cause-Code
   column of the following table.  The cause-name contains the
   corresponding token selected from the Cause-Name column.

   completion-cause         =  "Completion-Cause" ":" cause-code SP
                               cause-name CRLF
   cause-code               =  3DIGIT
   cause-name               =  *VCHAR


















Burnett & Shanmugham         Standards Track                   [Page 80]

RFC 6787                         MRCPv2                    November 2012


   +------------+-----------------------+------------------------------+
   | Cause-Code | Cause-Name            | Description                  |
   +------------+-----------------------+------------------------------+
   | 000        | success               | RECOGNIZE completed with a   |
   |            |                       | match or DEFINE-GRAMMAR      |
   |            |                       | succeeded in downloading and |
   |            |                       | compiling the grammar.       |
   |            |                       |                              |
   | 001        | no-match              | RECOGNIZE completed, but no  |
   |            |                       | match was found.             |
   |            |                       |                              |
   | 002        | no-input-timeout      | RECOGNIZE completed without  |
   |            |                       | a match due to a             |
   |            |                       | no-input-timeout.            |
   |            |                       |                              |
   | 003        | hotword-maxtime       | RECOGNIZE in hotword mode    |
   |            |                       | completed without a match    |
   |            |                       | due to a                     |
   |            |                       | recognition-timeout.         |
   |            |                       |                              |
   | 004        | grammar-load-failure  | RECOGNIZE failed due to      |
   |            |                       | grammar load failure.        |
   |            |                       |                              |
   | 005        | grammar-compilation-  | RECOGNIZE failed due to      |
   |            | failure               | grammar compilation failure. |
   |            |                       |                              |
   | 006        | recognizer-error      | RECOGNIZE request terminated |
   |            |                       | prematurely due to a         |
   |            |                       | recognizer error.            |
   |            |                       |                              |
   | 007        | speech-too-early      | RECOGNIZE request terminated |
   |            |                       | because speech was too       |
   |            |                       | early. This happens when the |
   |            |                       | audio stream is already      |
   |            |                       | "in-speech" when the         |
   |            |                       | RECOGNIZE request was        |
   |            |                       | received.                    |
   |            |                       |                              |
   | 008        | success-maxtime       | RECOGNIZE request terminated |
   |            |                       | because speech was too long  |
   |            |                       | but whatever was spoken till |
   |            |                       | that point was a full match. |
   |            |                       |                              |
   | 009        | uri-failure           | Failure accessing a URI.     |
   |            |                       |                              |
   | 010        | language-unsupported  | Language not supported.      |
   |            |                       |                              |




Burnett & Shanmugham         Standards Track                   [Page 81]

RFC 6787                         MRCPv2                    November 2012


   | 011        | cancelled             | A new RECOGNIZE cancelled    |
   |            |                       | this one, or a prior         |
   |            |                       | RECOGNIZE failed while this  |
   |            |                       | one was still in the queue.  |
   |            |                       |                              |
   | 012        | semantics-failure     | Recognition succeeded, but   |
   |            |                       | semantic interpretation of   |
   |            |                       | the recognized input failed. |
   |            |                       | The RECOGNITION-COMPLETE     |
   |            |                       | event MUST contain the       |
   |            |                       | Recognition result with only |
   |            |                       | input text and no            |
   |            |                       | interpretation.              |
   |            |                       |                              |
   | 013        | partial-match         | Speech Incomplete Timeout    |
   |            |                       | expired before there was a   |
   |            |                       | full match. But whatever was |
   |            |                       | spoken till that point was a |
   |            |                       | partial match to one or more |
   |            |                       | grammars.                    |
   |            |                       |                              |
   | 014        | partial-match-maxtime | The Recognition-Timeout      |
   |            |                       | expired before full match    |
   |            |                       | was achieved. But whatever   |
   |            |                       | was spoken till that point   |
   |            |                       | was a partial match to one   |
   |            |                       | or more grammars.            |
   |            |                       |                              |
   | 015        | no-match-maxtime      | The Recognition-Timeout      |
   |            |                       | expired. Whatever was spoken |
   |            |                       | till that point did not      |
   |            |                       | match any of the grammars.   |
   |            |                       | This cause could also be     |
   |            |                       | returned if the recognizer   |
   |            |                       | does not support detecting   |
   |            |                       | partial grammar matches.     |
   |            |                       |                              |
   | 016        | grammar-definition-   | Any DEFINE-GRAMMAR error     |
   |            | failure               | other than                   |
   |            |                       | grammar-load-failure and     |
   |            |                       | grammar-compilation-failure. |
   +------------+-----------------------+------------------------------+









Burnett & Shanmugham         Standards Track                   [Page 82]

RFC 6787                         MRCPv2                    November 2012


9.4.12.  Completion-Reason



   This header field MAY be specified in a RECOGNITION-COMPLETE event
   coming from the recognizer resource to the client.  This contains the
   reason text behind the RECOGNIZE request completion.  The server uses
   this header field to communicate text describing the reason for the
   failure, such as the specific error encountered in parsing a grammar
   markup.

   The completion reason text is provided for client use in logs and for
   debugging and instrumentation purposes.  Clients MUST NOT interpret
   the completion reason text.

   completion-reason        =  "Completion-Reason" ":"
                               quoted-string CRLF

9.4.13.  Recognizer-Context-Block



   This header field MAY be sent as part of the SET-PARAMS or GET-PARAMS
   request.  If the GET-PARAMS method contains this header field with no
   value, then it is a request to the recognizer to return the
   recognizer context block.  The response to such a message MAY contain
   a recognizer context block as a typed media message body.  If the
   server returns a recognizer context block, the response MUST contain
   this header field and its value MUST match the Content-ID of the
   corresponding media block.

   If the SET-PARAMS method contains this header field, it MUST also
   contain a message body containing the recognizer context data and a
   Content-ID matching this header field value.  This Content-ID MUST
   match the Content-ID that came with the context data during the
   GET-PARAMS operation.

   An implementation choosing to use this mechanism to hand off
   recognizer context data between servers MUST distinguish its
   implementation-specific block of data by using an IANA-registered
   content type in the IANA Media Type vendor tree.

   recognizer-context-block  =  "Recognizer-Context-Block" ":"
                                [1*VCHAR] CRLF

9.4.14.  Start-Input-Timers



   This header field MAY be sent as part of the RECOGNIZE request.  A
   value of false tells the recognizer to start recognition but not to
   start the no-input timer yet.  The recognizer MUST NOT start the
   timers until the client sends a START-INPUT-TIMERS request to the
   recognizer.  This is useful in the scenario when the recognizer and



Burnett & Shanmugham         Standards Track                   [Page 83]

RFC 6787                         MRCPv2                    November 2012


   synthesizer engines are not part of the same session.  In such
   configurations, when a kill-on-barge-in prompt is being played (see
   Section 8.4.2), the client wants the RECOGNIZE request to be
   simultaneously active so that it can detect and implement kill-on-
   barge-in.  However, the recognizer SHOULD NOT start the no-input
   timers until the prompt is finished.  The default value is "true".

   start-input-timers  =  "Start-Input-Timers" ":" BOOLEAN CRLF

9.4.15.  Speech-Complete-Timeout



   This header field specifies the length of silence required following
   user speech before the speech recognizer finalizes a result (either
   accepting it or generating a no-match result).  The Speech-Complete-
   Timeout value applies when the recognizer currently has a complete
   match against an active grammar, and specifies how long the
   recognizer MUST wait for more input before declaring a match.  By
   contrast, the Speech-Incomplete-Timeout is used when the speech is an
   incomplete match to an active grammar.  The value is in milliseconds.

  speech-complete-timeout = "Speech-Complete-Timeout" ":" 1*19DIGIT CRLF

   A long Speech-Complete-Timeout value delays the result to the client
   and therefore makes the application's response to a user slow.  A
   short Speech-Complete-Timeout may lead to an utterance being broken
   up inappropriately.  Reasonable speech complete timeout values are
   typically in the range of 0.3 seconds to 1.0 seconds.  The value for
   this header field ranges from 0 to an implementation-specific maximum
   value.  The default value for this header field is implementation
   specific.  This header field MAY occur in RECOGNIZE, SET-PARAMS, or
   GET-PARAMS.

9.4.16.  Speech-Incomplete-Timeout



   This header field specifies the required length of silence following
   user speech after which a recognizer finalizes a result.  The
   incomplete timeout applies when the speech prior to the silence is an
   incomplete match of all active grammars.  In this case, once the
   timeout is triggered, the partial result is rejected (with a
   Completion-Cause of "partial-match").  The value is in milliseconds.
   The value for this header field ranges from 0 to an implementation-
   specific maximum value.  The default value for this header field is
   implementation specific.

   speech-incomplete-timeout = "Speech-Incomplete-Timeout" ":" 1*19DIGIT
                                CRLF





Burnett & Shanmugham         Standards Track                   [Page 84]

RFC 6787                         MRCPv2                    November 2012


   The Speech-Incomplete-Timeout also applies when the speech prior to
   the silence is a complete match of an active grammar, but where it is
   possible to speak further and still match the grammar.  By contrast,
   the Speech-Complete-Timeout is used when the speech is a complete
   match to an active grammar and no further spoken words can continue
   to represent a match.

   A long Speech-Incomplete-Timeout value delays the result to the
   client and therefore makes the application's response to a user slow.
   A short Speech-Incomplete-Timeout may lead to an utterance being
   broken up inappropriately.

   The Speech-Incomplete-Timeout is usually longer than the Speech-
   Complete-Timeout to allow users to pause mid-utterance (for example,
   to breathe).  This header field MAY occur in RECOGNIZE, SET-PARAMS,
   or GET-PARAMS.

9.4.17.  DTMF-Interdigit-Timeout



   This header field specifies the inter-digit timeout value to use when
   recognizing DTMF input.  The value is in milliseconds.  The value for
   this header field ranges from 0 to an implementation-specific maximum
   value.  The default value is 5 seconds.  This header field MAY occur
   in RECOGNIZE, SET-PARAMS, or GET-PARAMS.

  dtmf-interdigit-timeout = "DTMF-Interdigit-Timeout" ":" 1*19DIGIT CRLF

9.4.18.  DTMF-Term-Timeout



   This header field specifies the terminating timeout to use when
   recognizing DTMF input.  The DTMF-Term-Timeout applies only when no
   additional input is allowed by the grammar; otherwise, the
   DTMF-Interdigit-Timeout applies.  The value is in milliseconds.  The
   value for this header field ranges from 0 to an implementation-
   specific maximum value.  The default value is 10 seconds.  This
   header field MAY occur in RECOGNIZE, SET-PARAMS, or GET-PARAMS.

   dtmf-term-timeout        =  "DTMF-Term-Timeout" ":" 1*19DIGIT CRLF

9.4.19.  DTMF-Term-Char



   This header field specifies the terminating DTMF character for DTMF
   input recognition.  The default value is NULL, which is indicated by
   an empty header field value.  This header field MAY occur in
   RECOGNIZE, SET-PARAMS, or GET-PARAMS.

   dtmf-term-char           =  "DTMF-Term-Char" ":" VCHAR CRLF




Burnett & Shanmugham         Standards Track                   [Page 85]

RFC 6787                         MRCPv2                    November 2012


9.4.20.  Failed-URI



   When a recognizer needs to fetch or access a URI and the access
   fails, the server SHOULD provide the failed URI in this header field
   in the method response, unless there are multiple URI failures, in
   which case one of the failed URIs MUST be provided in this header
   field in the method response.

   failed-uri               =  "Failed-URI" ":" absoluteURI CRLF

9.4.21.  Failed-URI-Cause



   When a recognizer method needs a recognizer to fetch or access a URI
   and the access fails, the server MUST provide the URI-specific or
   protocol-specific response code for the URI in the Failed-URI header
   field through this header field in the method response.  The value
   encoding is UTF-8 (RFC 3629 [RFC3629]) to accommodate any access
   protocol, some of which might have a response string instead of a
   numeric response code.

   failed-uri-cause         =  "Failed-URI-Cause" ":" 1*UTFCHAR CRLF

9.4.22.  Save-Waveform



   This header field allows the client to request the recognizer
   resource to save the audio input to the recognizer.  The recognizer
   resource MUST then attempt to record the recognized audio, without
   endpointing, and make it available to the client in the form of a URI
   returned in the Waveform-URI header field in the RECOGNITION-COMPLETE
   event.  If there was an error in recording the stream or the audio
   content is otherwise not available, the recognizer MUST return an
   empty Waveform-URI header field.  The default value for this field is
   "false".  This header field MAY occur in RECOGNIZE, SET-PARAMS, or
   GET-PARAMS.  See the discussion on the sensitivity of saved waveforms
   in Section 12.

   save-waveform            =  "Save-Waveform" ":" BOOLEAN CRLF

9.4.23.  New-Audio-Channel



   This header field MAY be specified in a RECOGNIZE request and allows
   the client to tell the server that, from this point on, further input
   audio comes from a different audio source, channel, or speaker.  If
   the recognizer resource had collected any input statistics or
   adaptation state, the recognizer resource MUST do what is appropriate
   for the specific recognition technology, which includes but is not
   limited to discarding any collected input statistics or adaptation
   state before starting the RECOGNIZE request.  Note that if there are



Burnett & Shanmugham         Standards Track                   [Page 86]

RFC 6787                         MRCPv2                    November 2012


   multiple resources that are sharing a media stream and are collecting
   or using this data, and the client issues this header field to one of
   the resources, the reset operation applies to all resources that use
   the shared media stream.  This helps in a number of use cases,
   including where the client wishes to reuse an open recognition
   session with an existing media session for multiple telephone calls.

   new-audio-channel        =  "New-Audio-Channel" ":" BOOLEAN
                               CRLF

9.4.24.  Speech-Language



   This header field specifies the language of recognition grammar data
   within a session or request, if it is not specified within the data.
   The value of this header field MUST follow RFC 5646 [RFC5646] for its
   values.  This MAY occur in DEFINE-GRAMMAR, RECOGNIZE, SET-PARAMS, or
   GET-PARAMS requests.

   speech-language          =  "Speech-Language" ":" 1*VCHAR CRLF

9.4.25.  Ver-Buffer-Utterance



   This header field lets the client request the server to buffer the
   utterance associated with this recognition request into a buffer
   available to a co-resident verifier resource.  The buffer is shared
   across resources within a session and is allocated when a verifier
   resource is added to this session.  The client MUST NOT send this
   header field unless a verifier resource is instantiated for the
   session.  The buffer is released when the verifier resource is
   released from the session.

9.4.26.  Recognition-Mode



   This header field specifies what mode the RECOGNIZE method will
   operate in.  The value choices are "normal" or "hotword".  If the
   value is "normal", the RECOGNIZE starts matching speech and DTMF to
   the grammars specified in the RECOGNIZE request.  If any portion of
   the speech does not match the grammar, the RECOGNIZE command
   completes with a no-match status.  Timers may be active to detect
   speech in the audio (see Section 9.4.14), so the RECOGNIZE method may
   complete because of a timeout waiting for speech.  If the value of
   this header field is "hotword", the RECOGNIZE method operates in
   hotword mode, where it only looks for the particular keywords or DTMF








Burnett & Shanmugham         Standards Track                   [Page 87]

RFC 6787                         MRCPv2                    November 2012


   sequences specified in the grammar and ignores silence or other
   speech in the audio stream.  The default value for this header field
   is "normal".  This header field MAY occur on the RECOGNIZE method.

   recognition-mode         =  "Recognition-Mode" ":"
                               "normal" / "hotword" CRLF

9.4.27.  Cancel-If-Queue



   This header field specifies what will happen if the client attempts
   to invoke another RECOGNIZE method when this RECOGNIZE request is
   already in progress for the resource.  The value for this header
   field is a Boolean.  A value of "true" means the server MUST
   terminate this RECOGNIZE request, with a Completion-Cause of
   "cancelled", if the client issues another RECOGNIZE request for the
   same resource.  A value of "false" for this header field indicates to
   the server that this RECOGNIZE request will continue to completion,
   and if the client issues more RECOGNIZE requests to the same
   resource, they are queued.  When the currently active RECOGNIZE
   request is stopped or completes with a successful match, the first
   RECOGNIZE method in the queue becomes active.  If the current
   RECOGNIZE fails, all RECOGNIZE methods in the pending queue are
   cancelled, and each generates a RECOGNITION-COMPLETE event with a
   Completion-Cause of "cancelled".  This header field MUST be present
   in every RECOGNIZE request.  There is no default value.

   cancel-if-queue          =  "Cancel-If-Queue" ":" BOOLEAN CRLF

9.4.28.  Hotword-Max-Duration



   This header field MAY be sent in a hotword mode RECOGNIZE request.
   It specifies the maximum length of an utterance (in seconds) that
   will be considered for hotword recognition.  This header field, along
   with Hotword-Min-Duration, can be used to tune performance by
   preventing the recognizer from evaluating utterances that are too
   short or too long to be one of the hotwords in the grammar(s).  The
   value is in milliseconds.  The default is implementation dependent.
   If present in a RECOGNIZE request specifying a mode other than
   "hotword", the header field is ignored.

   hotword-max-duration     =  "Hotword-Max-Duration" ":" 1*19DIGIT
                               CRLF

9.4.29.  Hotword-Min-Duration



   This header field MAY be sent in a hotword mode RECOGNIZE request.
   It specifies the minimum length of an utterance (in seconds) that
   will be considered for hotword recognition.  This header field, along



Burnett & Shanmugham         Standards Track                   [Page 88]

RFC 6787                         MRCPv2                    November 2012


   with Hotword-Max-Duration, can be used to tune performance by
   preventing the recognizer from evaluating utterances that are too
   short or too long to be one of the hotwords in the grammar(s).  The
   value is in milliseconds.  The default value is implementation
   dependent.  If present in a RECOGNIZE request specifying a mode other
   than "hotword", the header field is ignored.

   hotword-min-duration     =  "Hotword-Min-Duration" ":" 1*19DIGIT CRLF

9.4.30.  Interpret-Text



   The value of this header field is used to provide a pointer to the
   text for which a natural language interpretation is desired.  The
   value is either a URI or text.  If the value is a URI, it MUST be a
   Content-ID that refers to an entity of type 'text/plain' in the body
   of the message.  Otherwise, the server MUST treat the value as the
   text to be interpreted.  This header field MUST be used when invoking
   the INTERPRET method.

   interpret-text           =  "Interpret-Text" ":" 1*VCHAR CRLF

9.4.31.  DTMF-Buffer-Time



   This header field MAY be specified in a GET-PARAMS or SET-PARAMS
   method and is used to specify the amount of time, in milliseconds, of
   the type-ahead buffer for the recognizer.  This is the buffer that
   collects DTMF digits as they are pressed even when there is no
   RECOGNIZE command active.  When a subsequent RECOGNIZE method is
   received, it MUST look to this buffer to match the RECOGNIZE request.
   If the digits in the buffer are not sufficient, then it can continue
   to listen to more digits to match the grammar.  The default size of
   this DTMF buffer is platform specific.

   dtmf-buffer-time  =  "DTMF-Buffer-Time" ":" 1*19DIGIT CRLF

9.4.32.  Clear-DTMF-Buffer



   This header field MAY be specified in a RECOGNIZE method and is used
   to tell the recognizer to clear the DTMF type-ahead buffer before
   starting the RECOGNIZE.  The default value of this header field is
   "false", which does not clear the type-ahead buffer before starting
   the RECOGNIZE method.  If this header field is specified to be
   "true", then the RECOGNIZE will clear the DTMF buffer before starting
   recognition.  This means digits pressed by the caller before the
   RECOGNIZE command was issued are discarded.

   clear-dtmf-buffer  = "Clear-DTMF-Buffer" ":" BOOLEAN CRLF




Burnett & Shanmugham         Standards Track                   [Page 89]

RFC 6787                         MRCPv2                    November 2012


9.4.33.  Early-No-Match



   This header field MAY be specified in a RECOGNIZE method and is used
   to tell the recognizer that it MUST NOT wait for the end of speech
   before processing the collected speech to match active grammars.  A
   value of "true" indicates the recognizer MUST do early matching.  The
   default value for this header field if not specified is "false".  If
   the recognizer does not support the processing of the collected audio
   before the end of speech, this header field can be safely ignored.

   early-no-match  = "Early-No-Match" ":" BOOLEAN CRLF

9.4.34.  Num-Min-Consistent-Pronunciations



   This header field MAY be specified in a START-PHRASE-ENROLLMENT,
   SET-PARAMS, or GET-PARAMS method and is used to specify the minimum
   number of consistent pronunciations that must be obtained to voice
   enroll a new phrase.  The minimum value is 1.  The default value is
   implementation specific and MAY be greater than 1.

   num-min-consistent-pronunciations  =
                 "Num-Min-Consistent-Pronunciations" ":" 1*19DIGIT CRLF

9.4.35.  Consistency-Threshold



   This header field MAY be sent as part of the START-PHRASE-ENROLLMENT,
   SET-PARAMS, or GET-PARAMS method.  Used during voice enrollment, this
   header field specifies how similar to a previously enrolled
   pronunciation of the same phrase an utterance needs to be in order to
   be considered "consistent".  The higher the threshold, the closer the
   match between an utterance and previous pronunciations must be for
   the pronunciation to be considered consistent.  The range for this
   threshold is a float value between 0.0 and 1.0.  The default value
   for this header field is implementation specific.

   consistency-threshold    =  "Consistency-Threshold" ":" FLOAT CRLF

9.4.36.  Clash-Threshold



   This header field MAY be sent as part of the START-PHRASE-ENROLLMENT,
   SET-PARAMS, or GET-PARAMS method.  Used during voice enrollment, this
   header field specifies how similar the pronunciations of two
   different phrases can be before they are considered to be clashing.
   For example, pronunciations of phrases such as "John Smith" and "Jon
   Smits" may be so similar that they are difficult to distinguish
   correctly.  A smaller threshold reduces the number of clashes
   detected.  The range for this threshold is a float value between 0.0




Burnett & Shanmugham         Standards Track                   [Page 90]

RFC 6787                         MRCPv2                    November 2012


   and 1.0.  The default value for this header field is implementation
   specific.  Clash testing can be turned off completely by setting the
   Clash-Threshold header field value to 0.

   clash-threshold          =  "Clash-Threshold" ":" FLOAT CRLF

9.4.37.  Personal-Grammar-URI



   This header field specifies the speaker-trained grammar to be used or
   referenced during enrollment operations.  Phrases are added to this
   grammar during enrollment.  For example, a contact list for user
   "Jeff" could be stored at the Personal-Grammar-URI
   "http://myserver.example.com/myenrollmentdb/jeff-list".  The
   generated grammar syntax MAY be implementation specific.  There is no
   default value for this header field.  This header field MAY be sent
   as part of the START-PHRASE-ENROLLMENT, SET-PARAMS, or GET-PARAMS
   method.

   personal-grammar-uri     =  "Personal-Grammar-URI" ":" uri CRLF

9.4.38.  Enroll-Utterance



   This header field MAY be specified in the RECOGNIZE method.  If this
   header field is set to "true" and an Enrollment is active, the
   RECOGNIZE command MUST add the collected utterance to the personal
   grammar that is being enrolled.  The way in which this occurs is
   engine specific and may be an area of future standardization.  The
   default value for this header field is "false".

   enroll-utterance     =  "Enroll-Utterance" ":" BOOLEAN CRLF

9.4.39.  Phrase-Id



   This header field in a request identifies a phrase in an existing
   personal grammar for which enrollment is desired.  It is also
   returned to the client in the RECOGNIZE complete event.  This header
   field MAY occur in START-PHRASE-ENROLLMENT, MODIFY-PHRASE, or DELETE-
   PHRASE requests.  There is no default value for this header field.

   phrase-id                =  "Phrase-ID" ":" 1*VCHAR CRLF











Burnett & Shanmugham         Standards Track                   [Page 91]

RFC 6787                         MRCPv2                    November 2012


9.4.40.  Phrase-NL



   This string specifies the interpreted text to be returned when the
   phrase is recognized.  This header field MAY occur in START-PHRASE-
   ENROLLMENT and MODIFY-PHRASE requests.  There is no default value for
   this header field.

   phrase-nl                =  "Phrase-NL" ":" 1*UTFCHAR CRLF

9.4.41.  Weight



   The value of this header field represents the occurrence likelihood
   of a phrase in an enrolled grammar.  When using grammar enrollment,
   the system is essentially constructing a grammar segment consisting
   of a list of possible match phrases.  This can be thought of to be
   similar to the dynamic construction of a <one-of> tag in the W3C
   grammar specification.  Each enrolled-phrase becomes an item in the
   list that can be matched against spoken input similar to the <item>
   within a <one-of> list.  This header field allows you to assign a
   weight to the phrase (i.e., <item> entry) in the <one-of> list that
   is enrolled.  Grammar weights are normalized to a sum of one at
   grammar compilation time, so a weight value of 1 for each phrase in
   an enrolled grammar list indicates all items in that list have the
   same weight.  This header field MAY occur in START-PHRASE-ENROLLMENT
   and MODIFY-PHRASE requests.  The default value for this header field
   is implementation specific.

   weight                   =  "Weight" ":" FLOAT CRLF

9.4.42.  Save-Best-Waveform



   This header field allows the client to request the recognizer
   resource to save the audio stream for the best repetition of the
   phrase that was used during the enrollment session.  The recognizer
   MUST attempt to record the recognized audio and make it available to
   the client in the form of a URI returned in the Waveform-URI header
   field in the response to the END-PHRASE-ENROLLMENT method.  If there
   was an error in recording the stream or the audio data is otherwise
   not available, the recognizer MUST return an empty Waveform-URI
   header field.  This header field MAY occur in the START-PHRASE-
   ENROLLMENT, SET-PARAMS, and GET-PARAMS methods.

   save-best-waveform  =  "Save-Best-Waveform" ":" BOOLEAN CRLF








Burnett & Shanmugham         Standards Track                   [Page 92]

RFC 6787                         MRCPv2                    November 2012


9.4.43.  New-Phrase-Id



   This header field replaces the ID used to identify the phrase in a
   personal grammar.  The recognizer returns the new ID when using an
   enrollment grammar.  This header field MAY occur in MODIFY-PHRASE
   requests.

   new-phrase-id            =  "New-Phrase-ID" ":" 1*VCHAR CRLF

9.4.44.  Confusable-Phrases-URI



   This header field specifies a grammar that defines invalid phrases
   for enrollment.  For example, typical applications do not allow an
   enrolled phrase that is also a command word.  This header field MAY
   occur in RECOGNIZE requests that are part of an enrollment session.

   confusable-phrases-uri   =  "Confusable-Phrases-URI" ":" uri CRLF

9.4.45.  Abort-Phrase-Enrollment



   This header field MAY be specified in the END-PHRASE-ENROLLMENT
   method to abort the phrase enrollment, rather than committing the
   phrase to the personal grammar.

   abort-phrase-enrollment  =  "Abort-Phrase-Enrollment" ":"
                               BOOLEAN CRLF

9.5.  Recognizer Message Body



   A recognizer message can carry additional data associated with the
   request, response, or event.  The client MAY provide the grammar to
   be recognized in DEFINE-GRAMMAR or RECOGNIZE requests.  When one or
   more grammars are specified using the DEFINE-GRAMMAR method, the
   server MUST attempt to fetch, compile, and optimize the grammar
   before returning a response to the DEFINE-GRAMMAR method.  A
   RECOGNIZE request MUST completely specify the grammars to be active
   during the recognition operation, except when the RECOGNIZE method is
   being used to enroll a grammar.  During grammar enrollment, such
   grammars are OPTIONAL.  The server resource sends the recognition
   results in the RECOGNITION-COMPLETE event and the GET-RESULT
   response.  Grammars and recognition results are carried in the
   message body of the corresponding MRCPv2 messages.

9.5.1.  Recognizer Grammar Data



   Recognizer grammar data from the client to the server can be provided
   inline or by reference.  Either way, grammar data is carried as typed
   media entities in the message body of the RECOGNIZE or DEFINE-GRAMMAR



Burnett & Shanmugham         Standards Track                   [Page 93]

RFC 6787                         MRCPv2                    November 2012


   request.  All MRCPv2 servers MUST accept grammars in the XML form
   (media type 'application/srgs+xml') of the W3C's XML-based Speech
   Grammar Markup Format (SRGS) [W3C.REC-speech-grammar-20040316] and
   MAY accept grammars in other formats.  Examples include but are not
   limited to:

   o  the ABNF form (media type 'application/srgs') of SRGS

   o  Sun's Java Speech Grammar Format (JSGF)
      [refs.javaSpeechGrammarFormat]

   Additionally, MRCPv2 servers MAY support the Semantic Interpretation
   for Speech Recognition (SISR)
   [W3C.REC-semantic-interpretation-20070405] specification.

   When a grammar is specified inline in the request, the client MUST
   provide a Content-ID for that grammar as part of the content header
   fields.  If there is no space on the server to store the inline
   grammar, the request MUST return with a Completion-Cause code of 016
   "grammar-definition-failure".  Otherwise, the server MUST associate
   the inline grammar block with that Content-ID and MUST store it on
   the server for the duration of the session.  However, if the
   Content-ID is redefined later in the session through a subsequent
   DEFINE-GRAMMAR, the inline grammar previously associated with the
   Content-ID MUST be freed.  If the Content-ID is redefined through a
   subsequent DEFINE-GRAMMAR with an empty message body (i.e., no
   grammar definition), then in addition to freeing any grammar
   previously associated with the Content-ID, the server MUST clear all
   bindings and associations to the Content-ID.  Unless and until
   subsequently redefined, this URI MUST be interpreted by the server as
   one that has never been set.

   Grammars that have been associated with a Content-ID can be
   referenced through the 'session' URI scheme (see Section 13.6).  For
   example:
   session:help@root-level.store

   Grammar data MAY be specified using external URI references.  To do
   so, the client uses a body of media type 'text/uri-list' (see RFC
   2483 [RFC2483] ) to list the one or more URIs that point to the
   grammar data.  The client can use a body of media type 'text/
   grammar-ref-list' (see Section 13.5.1) if it wants to assign weights
   to the list of grammar URI.  All MRCPv2 servers MUST support grammar
   access using the 'http' and 'https' URI schemes.

   If the grammar data the client wishes to be used on a request
   consists of a mix of URI and inline grammar data, the client uses the
   'multipart/mixed' media type to enclose the 'text/uri-list',



Burnett & Shanmugham         Standards Track                   [Page 94]

RFC 6787                         MRCPv2                    November 2012


   'application/srgs', or 'application/srgs+xml' content entities.  The
   character set and encoding used in the grammar data are specified
   using to standard media type definitions.

   When more than one grammar URI or inline grammar block is specified
   in a message body of the RECOGNIZE request, the server interprets
   this as a list of grammar alternatives to match against.

   Content-Type:application/srgs+xml
   Content-ID:<request1@form-level.store>
   Content-Length:...

   <?xml version="1.0"?>

   <!-- the default grammar language is US English -->
   <grammar xmlns="http://www.w3.org/2001/06/grammar"
            xml:lang="en-US" version="1.0" root="request">

   <!-- single language attachment to tokens -->
         <rule id="yes">
               <one-of>
                     <item xml:lang="fr-CA">oui</item>
                     <item xml:lang="en-US">yes</item>
               </one-of>
         </rule>

   <!-- single language attachment to a rule expansion -->
         <rule id="request">
               may I speak to
               <one-of xml:lang="fr-CA">
                     <item>Michel Tremblay</item>
                     <item>Andre Roy</item>
               </one-of>
         </rule>

         <!-- multiple language attachment to a token -->
         <rule id="people1">
               <token lexicon="en-US,fr-CA"> Robert </token>
         </rule>












Burnett & Shanmugham         Standards Track                   [Page 95]

RFC 6787                         MRCPv2                    November 2012


         <!-- the equivalent single-language attachment expansion -->
         <rule id="people2">
               <one-of>
                     <item xml:lang="en-US">Robert</item>
                     <item xml:lang="fr-CA">Robert</item>
               </one-of>
         </rule>

         </grammar>

                           SRGS Grammar Example


   Content-Type:text/uri-list
   Content-Length:...

   session:help@root-level.store
   http://www.example.com/Directory-Name-List.grxml
   http://www.example.com/Department-List.grxml
   http://www.example.com/TAC-Contact-List.grxml
   session:menu1@menu-level.store

                         Grammar Reference Example


   Content-Type:multipart/mixed; boundary="break"

   --break
   Content-Type:text/uri-list
   Content-Length:...

   http://www.example.com/Directory-Name-List.grxml
   http://www.example.com/Department-List.grxml
   http://www.example.com/TAC-Contact-List.grxml

   --break
   Content-Type:application/srgs+xml
   Content-ID:<request1@form-level.store>
   Content-Length:...












Burnett & Shanmugham         Standards Track                   [Page 96]

RFC 6787                         MRCPv2                    November 2012


   <?xml version="1.0"?>

   <!-- the default grammar language is US English -->
   <grammar xmlns="http://www.w3.org/2001/06/grammar"
            xml:lang="en-US" version="1.0">

   <!-- single language attachment to tokens -->
         <rule id="yes">
               <one-of>
                     <item xml:lang="fr-CA">oui</item>
                     <item xml:lang="en-US">yes</item>
               </one-of>
         </rule>

   <!-- single language attachment to a rule expansion -->
         <rule id="request">
               may I speak to
               <one-of xml:lang="fr-CA">
                     <item>Michel Tremblay</item>
                     <item>Andre Roy</item>
               </one-of>
         </rule>

         <!-- multiple language attachment to a token -->
         <rule id="people1">
               <token lexicon="en-US,fr-CA"> Robert </token>
         </rule>

         <!-- the equivalent single-language attachment expansion -->
         <rule id="people2">
               <one-of>
                     <item xml:lang="en-US">Robert</item>
                     <item xml:lang="fr-CA">Robert</item>
               </one-of>
         </rule>

         </grammar>
   --break--

                      Mixed Grammar Reference Example

9.5.2.  Recognizer Result Data



   Recognition results are returned to the client in the message body of
   the RECOGNITION-COMPLETE event or the GET-RESULT response message as
   described in Section 6.3.  Element and attribute descriptions for the
   recognition portion of the NLSML format are provided in Section 9.6
   with a normative definition of the schema in Section 16.1.



Burnett & Shanmugham         Standards Track                   [Page 97]

RFC 6787                         MRCPv2                    November 2012


   Content-Type:application/nlsml+xml
   Content-Length:...

   <?xml version="1.0"?>
   <result xmlns="urn:ietf:params:xml:ns:mrcpv2"
           xmlns:ex="http://www.example.com/example"
           grammar="http://www.example.com/theYesNoGrammar">
       <interpretation>
           <instance>
                   <ex:response>yes</ex:response>
           </instance>
           <input>OK</input>
       </interpretation>
   </result>

                              Result Example

9.5.3.  Enrollment Result Data



   Enrollment results are returned to the client in the message body of
   the RECOGNITION-COMPLETE event as described in Section 6.3.  Element
   and attribute descriptions for the enrollment portion of the NLSML
   format are provided in Section 9.7 with a normative definition of the
   schema in Section 16.2.

9.5.4.  Recognizer Context Block



   When a client changes servers while operating on the behalf of the
   same incoming communication session, this header field allows the
   client to collect a block of opaque data from one server and provide
   it to another server.  This capability is desirable if the client
   needs different language support or because the server issued a
   redirect.  Here, the first recognizer resource may have collected
   acoustic and other data during its execution of recognition methods.
   After a server switch, communicating this data may allow the
   recognizer resource on the new server to provide better recognition.
   This block of data is implementation specific and MUST be carried as
   media type 'application/octets' in the body of the message.

   This block of data is communicated in the SET-PARAMS and GET-PARAMS
   method/response messages.  In the GET-PARAMS method, if an empty
   Recognizer-Context-Block header field is present, then the recognizer
   SHOULD return its vendor-specific context block, if any, in the
   message body as an entity of media type 'application/octets' with a
   specific Content-ID.  The Content-ID value MUST also be specified in
   the Recognizer-Context-Block header field in the GET-PARAMS response.
   The SET-PARAMS request wishing to provide this vendor-specific data
   MUST send it in the message body as a typed entity with the same



Burnett & Shanmugham         Standards Track                   [Page 98]

RFC 6787                         MRCPv2                    November 2012


   Content-ID that it received from the GET-PARAMS.  The Content-ID MUST
   also be sent in the Recognizer-Context-Block header field of the
   SET-PARAMS message.

   Each speech recognition implementation choosing to use this mechanism
   to hand off recognizer context data among servers MUST distinguish
   its implementation-specific block of data from other implementations
   by choosing a Content-ID that is recognizable among the participating
   servers and unlikely to collide with values chosen by another
   implementation.

9.6.  Recognizer Results



   The recognizer portion of NLSML (see Section 6.3.1) represents
   information automatically extracted from a user's utterances by a
   semantic interpretation component, where "utterance" is to be taken
   in the general sense of a meaningful user input in any modality
   supported by the MRCPv2 implementation.

9.6.1.  Markup Functions



   MRCPv2 recognizer resources employ the Natural Language Semantics
   Markup Language (NLSML) to interpret natural language speech input
   and to format the interpretation for consumption by an MRCPv2 client.

   The elements of the markup fall into the following general functional
   categories: interpretation, side information, and multi-modal
   integration.

9.6.1.1.  Interpretation



   Elements and attributes represent the semantics of a user's
   utterance, including the <result>, <interpretation>, and <instance>
   elements.  The <result> element contains the full result of
   processing one utterance.  It MAY contain multiple <interpretation>
   elements if the interpretation of the utterance results in multiple
   alternative meanings due to uncertainty in speech recognition or
   natural language understanding.  There are at least two reasons for
   providing multiple interpretations:

   1.  The client application might have additional information, for
       example, information from a database, that would allow it to
       select a preferred interpretation from among the possible
       interpretations returned from the semantic interpreter.







Burnett & Shanmugham         Standards Track                   [Page 99]

RFC 6787                         MRCPv2                    November 2012


   2.  A client-based dialog manager (e.g., VoiceXML
       [W3C.REC-voicexml20-20040316]) that was unable to select between
       several competing interpretations could use this information to
       go back to the user and find out what was intended.  For example,
       it could issue a SPEAK request to a synthesizer resource to emit
       "Did you say 'Boston' or 'Austin'?"

9.6.1.2.  Side Information



   These are elements and attributes representing additional information
   about the interpretation, over and above the interpretation itself.
   Side information includes:

   1.  Whether an interpretation was achieved (the <nomatch> element)
       and the system's confidence in an interpretation (the
       "confidence" attribute of <interpretation>).

   2.  Alternative interpretations (<interpretation>)

   3.  Input formats and Automatic Speech Recognition (ASR) information:
       the <input> element, representing the input to the semantic
       interpreter.

9.6.1.3.  Multi-Modal Integration



   When more than one modality is available for input, the
   interpretation of the inputs needs to be coordinated.  The "mode"
   attribute of <input> supports this by indicating whether the
   utterance was input by speech, DTMF, pointing, etc.  The "timestamp-
   start" and "timestamp-end" attributes of <input> also provide for
   temporal coordination by indicating when inputs occurred.

9.6.2.  Overview of Recognizer Result Elements and Their Relationships



   The recognizer elements in NLSML fall into two categories:

   1.  description of the input that was processed, and

   2.  description of the meaning which was extracted from the input.

   Next to each element are its attributes.  In addition, some elements
   can contain multiple instances of other elements.  For example, a
   <result> can contain multiple <interpretation> elements, each of
   which is taken to be an alternative.  Similarly, <input> can contain
   multiple child <input> elements, which are taken to be cumulative.
   To illustrate the basic usage of these elements, as a simple example,





Burnett & Shanmugham         Standards Track                  [Page 100]

RFC 6787                         MRCPv2                    November 2012


   consider the utterance "OK" (interpreted as "yes").  The example
   illustrates how that utterance and its interpretation would be
   represented in the NLSML markup.

   <?xml version="1.0"?>
   <result xmlns="urn:ietf:params:xml:ns:mrcpv2"
           xmlns:ex="http://www.example.com/example"
           grammar="http://www.example.com/theYesNoGrammar">
     <interpretation>
        <instance>
           <ex:response>yes</ex:response>
         </instance>
       <input>OK</input>
     </interpretation>
   </result>

   This example includes only the minimum required information.  There
   is an overall <result> element, which includes one interpretation and
   an input element.  The interpretation contains the application-
   specific element "<response>", which is the semantically interpreted
   result.

9.6.3.  Elements and Attributes



9.6.3.1.  <result> Root Element



   The root element of the markup is <result>.  The <result> element
   includes one or more <interpretation> elements.  Multiple
   interpretations can result from ambiguities in the input or in the
   semantic interpretation.  If the "grammar" attribute does not apply
   to all of the interpretations in the result, it can be overridden for
   individual interpretations at the <interpretation> level.

   Attributes:

   1.  grammar: The grammar or recognition rule matched by this result.
       The format of the grammar attribute will match the rule reference
       semantics defined in the grammar specification.  Specifically,
       the rule reference is in the external XML form for grammar rule
       references.  The markup interpreter needs to know the grammar
       rule that is matched by the utterance because multiple rules may
       be simultaneously active.  The value is the grammar URI used by
       the markup interpreter to specify the grammar.  The grammar can
       be overridden by a grammar attribute in the <interpretation>
       element if the input was ambiguous as to which grammar it
       matched.  If all interpretation elements within the result
       element contain their own grammar attributes, the attribute can
       be dropped from the result element.



Burnett & Shanmugham         Standards Track                  [Page 101]

RFC 6787                         MRCPv2                    November 2012


   <?xml version="1.0"?>
   <result xmlns="urn:ietf:params:xml:ns:mrcpv2"
           grammar="http://www.example.com/grammar">
     <interpretation>
      ....
     </interpretation>
   </result>

9.6.3.2.  <interpretation> Element



   An <interpretation> element contains a single semantic
   interpretation.

   Attributes:

   1.  confidence: A float value from 0.0-1.0 indicating the semantic
       analyzer's confidence in this interpretation.  A value of 1.0
       indicates maximum confidence.  The values are implementation
       dependent but are intended to align with the value interpretation
       for the confidence MRCPv2 header field defined in Section 9.4.1.
       This attribute is OPTIONAL.

   2.  grammar: The grammar or recognition rule matched by this
       interpretation (if needed to override the grammar specification
       at the <interpretation> level.)  This attribute is only needed
       under <interpretation> if it is necessary to override a grammar
       that was defined at the <result> level.  Note that the grammar
       attribute for the interpretation element is optional if and only
       if the grammar attribute is specified in the <result> element.

   Interpretations MUST be sorted best-first by some measure of
   "goodness".  The goodness measure is "confidence" if present;
   otherwise, it is some implementation-specific indication of quality.

   The grammar is expected to be specified most frequently at the
   <result> level.  However, it can be overridden at the
   <interpretation> level because it is possible that different
   interpretations may match different grammar rules.

   The <interpretation> element includes an optional <input> element
   containing the input being analyzed, and at least one <instance>
   element containing the interpretation of the utterance.

   <interpretation confidence="0.75"
                   grammar="http://www.example.com/grammar">
       ...
   </interpretation>




Burnett & Shanmugham         Standards Track                  [Page 102]

RFC 6787                         MRCPv2                    November 2012


9.6.3.3.  <instance> Element



   The <instance> element contains the interpretation of the utterance.
   When the Semantic Interpretation for Speech Recognition format is
   used, the <instance> element contains the XML serialization of the
   result using the approach defined in that specification.  When there
   is semantic markup in the grammar that does not create semantic
   objects, but instead only does a semantic translation of a portion of
   the input, such as translating "coke" to "coca-cola", the instance
   contains the whole input but with the translation applied.  The NLSML
   looks like the markup in Figure 2 below.  If there are no semantic
   objects created, nor any semantic translation, the instance value is
   the same as the input value.

   Attributes:

   1.  confidence: Each element of the instance MAY have a confidence
       attribute, defined in the NLSML namespace.  The confidence
       attribute contains a float value in the range from 0.0-1.0
       reflecting the system's confidence in the analysis of that slot.
       A value of 1.0 indicates maximum confidence.  The values are
       implementation dependent, but are intended to align with the
       value interpretation for the MRCPv2 header field Confidence-
       Threshold defined in Section 9.4.1.  This attribute is OPTIONAL.

   <instance>
     <nameAddress>
         <street confidence="0.75">123 Maple Street</street>
         <city>Mill Valley</city>
         <state>CA</state>
         <zip>90952</zip>
     </nameAddress>
   </instance>
   <input>
     My address is 123 Maple Street,
     Mill Valley, California, 90952
   </input>


   <instance>
       I would like to buy a coca-cola
   </instance>
   <input>
     I would like to buy a coke
   </input>

                          Figure 2: NSLML Example




Burnett & Shanmugham         Standards Track                  [Page 103]

RFC 6787                         MRCPv2                    November 2012


9.6.3.4.  <input> Element



   The <input> element is the text representation of a user's input.  It
   includes an optional "confidence" attribute, which indicates the
   recognizer's confidence in the recognition result (as opposed to the
   confidence in the interpretation, which is indicated by the
   "confidence" attribute of <interpretation>).  Optional "timestamp-
   start" and "timestamp-end" attributes indicate the start and end
   times of a spoken utterance, in ISO 8601 format [ISO.8601.1988].

   Attributes:

   1.  timestamp-start: The time at which the input began. (optional)

   2.  timestamp-end: The time at which the input ended. (optional)

   3.  mode: The modality of the input, for example, speech, DTMF, etc.
       (optional)

   4.  confidence: The confidence of the recognizer in the correctness
       of the input in the range 0.0 to 1.0. (optional)

   Note that it may not make sense for temporally overlapping inputs to
   have the same mode; however, this constraint is not expected to be
   enforced by implementations.

   When there is no time zone designator, ISO 8601 time representations
   default to local time.

   There are three possible formats for the <input> element.

   1.  The <input> element can contain simple text:

       <input>onions</input>

       A future possibility is for <input> to contain not only text but
       additional markup that represents prosodic information that was
       contained in the original utterance and extracted by the speech
       recognizer.  This depends on the availability of ASRs that are
       capable of producing prosodic information.  MRCPv2 clients MUST
       be prepared to receive such markup and MAY make use of it.

   2.  An <input> tag can also contain additional <input> tags.  Having
       additional input elements allows the representation to support
       future multi-modal inputs as well as finer-grained speech
       information, such as timestamps for individual words and word-
       level confidences.




Burnett & Shanmugham         Standards Track                  [Page 104]

RFC 6787                         MRCPv2                    November 2012


       <input>
            <input mode="speech" confidence="0.5"
                timestamp-start="2000-04-03T0:00:00"
                timestamp-end="2000-04-03T0:00:00.2">fried</input>
            <input mode="speech" confidence="1.0"
                timestamp-start="2000-04-03T0:00:00.25"
                timestamp-end="2000-04-03T0:00:00.6">onions</input>
       </input>

   3.  Finally, the <input> element can contain <nomatch> and <noinput>
       elements, which describe situations in which the speech
       recognizer received input that it was unable to process or did
       not receive any input at all, respectively.

9.6.3.5.  <nomatch> Element



   The <nomatch> element under <input> is used to indicate that the
   semantic interpreter was unable to successfully match any input with
   confidence above the threshold.  It can optionally contain the text
   of the best of the (rejected) matches.

   <interpretation>
      <instance/>
         <input confidence="0.1">
            <nomatch/>
         </input>
   </interpretation>
   <interpretation>
      <instance/>
      <input mode="speech" confidence="0.1">
        <nomatch>I want to go to New York</nomatch>
      </input>
   </interpretation>

9.6.3.6.  <noinput> Element



   <noinput> indicates that there was no input -- a timeout occurred in
   the speech recognizer due to silence.
   <interpretation>
      <instance/>
      <input>
         <noinput/>
      </input>
   </interpretation>

   If there are multiple levels of inputs, the most natural place for
   <nomatch> and <noinput> elements to appear is under the highest level
   of <input> for <noinput>, and under the appropriate level of



Burnett & Shanmugham         Standards Track                  [Page 105]

RFC 6787                         MRCPv2                    November 2012


   <interpretation> for <nomatch>.  So, <noinput> means "no input at
   all" and <nomatch> means "no match in speech modality" or "no match
   in DTMF modality".  For example, to represent garbled speech combined
   with DTMF "1 2 3 4", the markup would be:
   <input>
      <input mode="speech"><nomatch/></input>
      <input mode="dtmf">1 2 3 4</input>
   </input>

   Note: while <noinput> could be represented as an attribute of input,
   <nomatch> cannot, since it could potentially include PCDATA content
   with the best match.  For parallelism, <noinput> is also an element.

9.7.  Enrollment Results



   All enrollment elements are contained within a single
   <enrollment-result> element under <result>.  The elements are
   described below and have the schema defined in Section 16.2.  The
   following elements are defined:

   1.  num-clashes

   2.  num-good-repetitions

   3.  num-repetitions-still-needed

   4.  consistency-status

   5.  clash-phrase-ids

   6.  transcriptions

   7.  confusable-phrases

9.7.1.  <num-clashes> Element



   The <num-clashes> element contains the number of clashes that this
   pronunciation has with other pronunciations in an active enrollment
   session.  The associated Clash-Threshold header field determines the
   sensitivity of the clash measurement.  Note that clash testing can be
   turned off completely by setting the Clash-Threshold header field
   value to 0.

9.7.2.  <num-good-repetitions> Element



   The <num-good-repetitions> element contains the number of consistent
   pronunciations obtained so far in an active enrollment session.




Burnett & Shanmugham         Standards Track                  [Page 106]

RFC 6787                         MRCPv2                    November 2012


9.7.3.  <num-repetitions-still-needed> Element



   The <num-repetitions-still-needed> element contains the number of
   consistent pronunciations that must still be obtained before the new
   phrase can be added to the enrollment grammar.  The number of
   consistent pronunciations required is specified by the client in the
   request header field Num-Min-Consistent-Pronunciations.  The returned
   value must be 0 before the client can successfully commit a phrase to
   the grammar by ending the enrollment session.

9.7.4.  <consistency-status> Element



   The <consistency-status> element is used to indicate how consistent
   the repetitions are when learning a new phrase.  It can have the
   values of consistent, inconsistent, and undecided.

9.7.5.  <clash-phrase-ids> Element



   The <clash-phrase-ids> element contains the phrase IDs of clashing
   pronunciation(s), if any.  This element is absent if there are no
   clashes.

9.7.6.  <transcriptions> Element



   The <transcriptions> element contains the transcriptions returned in
   the last repetition of the phrase being enrolled.

9.7.7.  <confusable-phrases> Element



   The <confusable-phrases> element contains a list of phrases from a
   command grammar that are confusable with the phrase being added to
   the personal grammar.  This element MAY be absent if there are no
   confusable phrases.

9.8.  DEFINE-GRAMMAR



   The DEFINE-GRAMMAR method, from the client to the server, provides
   one or more grammars and requests the server to access, fetch, and
   compile the grammars as needed.  The DEFINE-GRAMMAR method
   implementation MUST do a fetch of all external URIs that are part of
   that operation.  If caching is implemented, this URI fetching MUST
   conform to the cache control hints and parameter header fields
   associated with the method in deciding whether the URIs should be
   fetched from cache or from the external server.  If these hints/
   parameters are not specified in the method, the values set for the
   session using SET-PARAMS/GET-PARAMS apply.  If it was not set for the
   session, their default values apply.




Burnett & Shanmugham         Standards Track                  [Page 107]

RFC 6787                         MRCPv2                    November 2012


   If the server resource is in the recognition state, the DEFINE-
   GRAMMAR request MUST respond with a failure status.

   If the resource is in the idle state and is able to successfully
   process the supplied grammars, the server MUST return a success code
   status and the request-state MUST be COMPLETE.

   If the recognizer resource could not define the grammar for some
   reason (for example, if the download failed, the grammar failed to
   compile, or the grammar was in an unsupported form), the MRCPv2
   response for the DEFINE-GRAMMAR method MUST contain a failure status-
   code of 407 and contain a Completion-Cause header field describing
   the failure reason.

   C->S:MRCP/2.0 ... DEFINE-GRAMMAR 543257
   Channel-Identifier:32AECB23433801@speechrecog
   Content-Type:application/srgs+xml
   Content-ID:<request1@form-level.store>
   Content-Length:...

   <?xml version="1.0"?>

   <!-- the default grammar language is US English -->
   <grammar xmlns="http://www.w3.org/2001/06/grammar"
            xml:lang="en-US" version="1.0">

   <!-- single language attachment to tokens -->
   <rule id="yes">
               <one-of>
                     <item xml:lang="fr-CA">oui</item>
                     <item xml:lang="en-US">yes</item>
               </one-of>
         </rule>

   <!-- single language attachment to a rule expansion -->
         <rule id="request">
               may I speak to
               <one-of xml:lang="fr-CA">
                     <item>Michel Tremblay</item>
                     <item>Andre Roy</item>
               </one-of>
         </rule>

         </grammar>

   S->C:MRCP/2.0 ... 543257 200 COMPLETE
   Channel-Identifier:32AECB23433801@speechrecog
           Completion-Cause:000 success



Burnett & Shanmugham         Standards Track                  [Page 108]

RFC 6787                         MRCPv2                    November 2012


   C->S:MRCP/2.0 ... DEFINE-GRAMMAR 543258
   Channel-Identifier:32AECB23433801@speechrecog
   Content-Type:application/srgs+xml
   Content-ID:<helpgrammar@root-level.store>
   Content-Length:...

   <?xml version="1.0"?>

   <!-- the default grammar language is US English -->
   <grammar xmlns="http://www.w3.org/2001/06/grammar"
            xml:lang="en-US" version="1.0">

         <rule id="request">
               I need help
         </rule>

   S->C:MRCP/2.0 ... 543258 200 COMPLETE
   Channel-Identifier:32AECB23433801@speechrecog
           Completion-Cause:000 success

   C->S:MRCP/2.0 ... DEFINE-GRAMMAR 543259
   Channel-Identifier:32AECB23433801@speechrecog
   Content-Type:application/srgs+xml
   Content-ID:<request2@field-level.store>
   Content-Length:...

   <?xml version="1.0" encoding="UTF-8"?>

   <!DOCTYPE grammar PUBLIC "-//W3C//DTD GRAMMAR 1.0//EN"
                     "http://www.w3.org/TR/speech-grammar/grammar.dtd">

   <grammar xmlns="http://www.w3.org/2001/06/grammar" xml:lang="en"
   xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
          xsi:schemaLocation="http://www.w3.org/2001/06/grammar
              http://www.w3.org/TR/speech-grammar/grammar.xsd"
              version="1.0" mode="voice" root="basicCmd">

   <meta name="author" content="Stephanie Williams"/>

   <rule id="basicCmd" scope="public">
     <example> please move the window </example>
     <example> open a file </example>

     <ruleref
       uri="http://grammar.example.com/politeness.grxml#startPolite"/>






Burnett & Shanmugham         Standards Track                  [Page 109]

RFC 6787                         MRCPv2                    November 2012


     <ruleref uri="#command"/>
     <ruleref
       uri="http://grammar.example.com/politeness.grxml#endPolite"/>
   </rule>

   <rule id="command">
     <ruleref uri="#action"/> <ruleref uri="#object"/>
   </rule>

   <rule id="action">
      <one-of>
         <item weight="10"> open   <tag>open</tag>   </item>
         <item weight="2">  close  <tag>close</tag>  </item>
         <item weight="1">  delete <tag>delete</tag> </item>
         <item weight="1">  move   <tag>move</tag>   </item>
      </one-of>
   </rule>

   <rule id="object">
     <item repeat="0-1">
       <one-of>
         <item> the </item>
         <item> a </item>
       </one-of>
     </item>

     <one-of>
         <item> window </item>
         <item> file </item>
         <item> menu </item>
     </one-of>
   </rule>

   </grammar>


   S->C:MRCP/2.0 ... 543259 200 COMPLETE
   Channel-Identifier:32AECB23433801@speechrecog
           Completion-Cause:000 success

   C->S:MRCP/2.0 ... RECOGNIZE 543260
   Channel-Identifier:32AECB23433801@speechrecog
           N-Best-List-Length:2
   Content-Type:text/uri-list
   Content-Length:...






Burnett & Shanmugham         Standards Track                  [Page 110]

RFC 6787                         MRCPv2                    November 2012


   session:request1@form-level.store
   session:request2@field-level.store
   session:helpgramar@root-level.store

   S->C:MRCP/2.0 ... 543260 200 IN-PROGRESS
   Channel-Identifier:32AECB23433801@speechrecog

   S->C:MRCP/2.0 ... START-OF-INPUT 543260 IN-PROGRESS
   Channel-Identifier:32AECB23433801@speechrecog

   S->C:MRCP/2.0 ... RECOGNITION-COMPLETE 543260 COMPLETE
   Channel-Identifier:32AECB23433801@speechrecog
   Completion-Cause:000 success
   Waveform-URI:<http://web.media.com/session123/audio.wav>;
                size=124535;duration=2340
   Content-Type:application/x-nlsml
   Content-Length:...

   <?xml version="1.0"?>
   <result xmlns="urn:ietf:params:xml:ns:mrcpv2"
           xmlns:ex="http://www.example.com/example"
           grammar="session:request1@form-level.store">
           <interpretation>
               <instance name="Person">
               <ex:Person>
                   <ex:Name> Andre Roy </ex:Name>
               </ex:Person>
            </instance>
            <input>   may I speak to Andre Roy </input>
       </interpretation>
   </result>

                          Define Grammar Example

9.9.  RECOGNIZE



   The RECOGNIZE method from the client to the server requests the
   recognizer to start recognition and provides it with one or more
   grammar references for grammars to match against the input media.
   The RECOGNIZE method can carry header fields to control the
   sensitivity, confidence level, and the level of detail in results
   provided by the recognizer.  These header field values override the
   current values set by a previous SET-PARAMS method.

   The RECOGNIZE method can request the recognizer resource to operate
   in normal or hotword mode as specified by the Recognition-Mode header
   field.  The default value is "normal".  If the resource could not
   start a recognition, the server MUST respond with a failure status-



Burnett & Shanmugham         Standards Track                  [Page 111]

RFC 6787                         MRCPv2                    November 2012


   code of 407 and a Completion-Cause header field in the response
   describing the cause of failure.

   The RECOGNIZE request uses the message body to specify the grammars
   applicable to the request.  The active grammar(s) for the request can
   be specified in one of three ways.  If the client needs to explicitly
   control grammar weights for the recognition operation, it MUST employ
   method 3 below.  The order of these grammars specifies the precedence
   of the grammars that is used when more than one grammar in the list
   matches the speech; in this case, the grammar with the higher
   precedence is returned as a match.  This precedence capability is
   useful in applications like VoiceXML browsers to order grammars
   specified at the dialog, document, and root level of a VoiceXML
   application.

   1.  The grammar MAY be placed directly in the message body as typed
       content.  If more than one grammar is included in the body, the
       order of inclusion controls the corresponding precedence for the
       grammars during recognition, with earlier grammars in the body
       having a higher precedence than later ones.

   2.  The body MAY contain a list of grammar URIs specified in content
       of media type 'text/uri-list' [RFC2483].  The order of the URIs
       determines the corresponding precedence for the grammars during
       recognition, with highest precedence first and decreasing for
       each URI thereaf