Hi,

I am trying to use two separate MRCP profiles in the context of the same dialplan.

The configurations work fine as long as a given dialplan only uses _one_ of the profiles during a given call.

However I currently I have a dialplan where I would like to use profile A. (connecting to MRCP server 1) to field one question, and profile B. (connecting to MRCP server 2) to field another question.

I issue the following command for profile A. and question 1:

%[current_grammar=s0]/usr/local/freeswitch/scripts/test-ivr/prompts/prompt.wav detect:unimrcp:profile-a {start-input-timers=false,no-input-timeout=3000,speech-incomplete-timeout=3500,recognition-timeout=5000,define-grammar=true,save-waveform=true,speech-language=es-MX}builtin:speech/grammar1

and the next one for profile B. and question 2:

%[current_grammar=s1]/usr/local/freeswitch/scripts/test-ivr/prompts/prompt.wav detect:unimrcp:profile-b {start-input-timers=false,no-input-timeout=3000,speech-incomplete-timeout=1500,recognition-timeout=5000,define-grammar=true,save-waveform=true}builtin:grammar/grammar2?language=es-MX

Again, these both work fine if I only use one profile for a given dialplan application or call. However, in the current case, I can see in my FreeSWITCH logs, in my ESL logs, and in the respective UniMRCP server logs, that the second request is apparently somehow reusing the session and profile details from the first request.

In my log I see the correct command is sent, but FreeSWITCH/the UniMRCP client still sends the second request using the connection details from the first profile, which then fails because grammar2 is not available via profile A [because it is handled by the server sitting behind profile B]. If I reverse the order, I see that both requests again arrive at the MRCP server attached to profile B.

I dug around the confluence wiki and found some references to parameters being sticky:

https://freeswitch.org/confluence/display/FREESWITCH/mod_dptools%3A+play_and_detect_speech#mod_dptools:play_and_detect_speech-Lua

and also tried manually setting "play_and_detect_speech_close_asr=true", but nothing seems to close the session.

Any hints would be greatly appreciated; alternatively if this impossible without digging around in the mod_ source that would be good to know too.

Best regards,

Joe