Voice Gateway Package MRCP Question

[xiaosen weng]

    I configure the 3945E voice gateway as a mrcp client to connect to the third party MRCP ASR Server(not nuance asr).

when the voice call setup. I capture the pcap in the mrcp server , from  the pcap, I see that the MRCP packages which is send from 3945e voice gateway is mixed, you can see the attach 1.png,  the frame not only contains MRCP "RECONGIZE" single, but also contains other data which is own to "DEFINE-GRAMMAR" single.

    My question is , does voice gateway has any command that can package the mrcp single more standard.  like the sip protocol,  every frame package only contains one single(for examle, invite data single in one frame, 180 ring data in one frame)

[xiaosen weng]

  the following picture is the MRCP ASR capture which is send by voice gateway.    my question is , why one Frame not started with the standard MRCP recongnize singal, but also contains other data? how cisco voice gateway organize the MRCP data to send?

[ptindall]

It's TCP/IP, it's a byte stream, it works that way.  The length of your MRCP message is the field following MRCP/2.0; that's what you need to use to find the end of message.

You might find it more useful in Wireshark to right-click on your packet and select Follow TCP Conversation to get the complete message flow across multiple TCP packets.

[xiaosen weng]

Dear Paul Tindall

let me have a example.

the MRCP message like follows. DEFINE-GRAMMAR and RECOGNIZE message

MRCP/2.0 289 DEFINE-GRAMMAR 1

Channel-Identifier: 7a2e74f23eb64ce4@speechrecog

Speech-Language: en-US

Content-Base: http://192.168.0.199:7000/CVP/

Content-Type: text/uri-list

Content-Id: field80@field.grammar

Content-Length: 46

MRCP/2.0 574 RECOGNIZE 3

Channel-Identifier: 7a2e74f23eb64ce4@speechrecog

Apr 23 13:31:54.130: Speech-Language: en-US

Confidence-Threshold: 0.40

Sensitivity-Level: 0.50

Speed-Vs-Accuracy: 0.50

Cancel-If-Queue: false

Dtmf-Interdigit-Timeout: 10000

Dtmf-Term-Timeout: 0

Dtmf-Term-Char: #

No-Input-Timeout: 8000

N-Best-List-Length: 1

Logging-Tag: 145970

Accept-Charset: utf-8

Content-Base: http://192.168.0.199:7000/CVP/

Media-Type: audio/basic

Start-Input-Timers: false

Apr 23 13:31:54.130: Content-Type: text/uri-list

Content-Length: 53

Apr 23 13:31:54.130: session:field80@field.grammar

session:help@grammar

and voice gateway use TCP/IP protocol to send to the MRCP server, but the voice gateway don't package one message one frame, it package the MRCP message like follows. it seems that frame 1 and frame 2 not an standard MRCP message

frame 1 data

MRCP/2.0 289 DEFINE-GRAMMAR 1

Channel-Identifier: 7a2e74f23eb64ce4@speechrecog

Speech-Language: en-US

Content-Base: http://192.168.0.199:7000/CVP/

Content-Type: text/

frame 2 data

uri-list

Content-Id: field80@field.grammar

Content-Length: 46

MRCP/2.0 574 RECOGNIZE 3

Channel-Identifier: 7a2e74f23eb64ce4@speechrecog

Apr 23 13:31:54.130: Speech-Language: en-US

Confidence-Threshold: 0.40

Sensitivity-Level: 0.50

Speed-Vs-Accuracy: 0.50

Cancel-If-Queue: false

Dtmf-Interdigit-Timeout: 10000

Dtmf-Term-Timeout: 0

Dtmf-Term-Char: #

No-Input-Timeout: 8000

N-Best-List-Length: 1

Logging-Tag: 145970

Accept-Charset: utf-8

Content-Base: http://192.168.0.199:7000/CVP/

Media-Type: audio/basic

Start-Input-Timers: false

Apr 23 13:31:54.130: Content-Type: text/uri-list

Content-Length: 53

Apr 23 13:31:54.130: session:field80@field.grammar

session:help@grammar

does voice gateway can support send the following type frame data ? (like the sip , one rame one message)

frame 1 data

MRCP/2.0 289 DEFINE-GRAMMAR 1

Channel-Identifier: 7a2e74f23eb64ce4@speechrecog

Speech-Language: en-US

Content-Base: http://192.168.0.199:7000/CVP/

Content-Type: text/uri-list

Content-Id: field80@field.grammar

Content-Length: 46

frame 2 data

MRCP/2.0 574 RECOGNIZE 3

Channel-Identifier: 7a2e74f23eb64ce4@speechrecog

Apr 23 13:31:54.130: Speech-Language: en-US

Confidence-Threshold: 0.40

Sensitivity-Level: 0.50

Speed-Vs-Accuracy: 0.50

Cancel-If-Queue: false

Dtmf-Interdigit-Timeout: 10000

Dtmf-Term-Timeout: 0

Dtmf-Term-Char: #

No-Input-Timeout: 8000

N-Best-List-Length: 1

Logging-Tag: 145970

Accept-Charset: utf-8

Content-Base: http://192.168.0.199:7000/CVP/

Media-Type: audio/basic

Start-Input-Timers: false

Apr 23 13:31:54.130: Content-Type: text/uri-list

Content-Length: 53

Apr 23 13:31:54.130: session:field80@field.grammar

session:help@grammar

Best Regards

翁小森

Sean

上海声浩信息科技股份有限公司

Shanghai Sunin Information Technology Ltd.,Co.

上海市徐汇区钦州北路1088号74栋718室 北京朝阳区建国路88号现代城SOHO 6号楼1606室

电话：021-64515198-8226 电话：010-62385035

传真：021-64825366 传真：010-65618823

手机：18616701182

mail：wengxiaosen@soundhao.com

[ptindall]

If you right-click on the TCP packet in Wireshark and view the TCP conversation to see the entire MRCP message flow, are you saying the MRCP messages are not MRCP protocol compliant?   Splitting of MRCP messages across network packets is entirely normal and something that has to be allowed for.  That's why there's a length field as the 2nd token in the MRCP message, to make decoding of the message stream simpler.

[xiaosen weng]

ok,thank you for your help.

Best Regards

翁小森

Sean

上海声浩信息科技股份有限公司

Shanghai Sunin Information Technology Ltd.,Co.

上海市徐汇区钦州北路1088号74栋718室 北京朝阳区建国路88号现代城SOHO 6号楼1606室

电话：021-64515198-8226 电话：010-62385035

传真：021-64825366 传真：010-65618823

手机：18616701182

mail：wengxiaosen@soundhao.com