I am working with freeswitch/LUA.  This is what I am using for my audio capture function:

play_and_detect_speech(select_an_option.wav detect:unimrcp:unimrcpv2-azure {start-input-timers = false, no-input-timeout = 10000,recognition-timeout = 20000, speech-incomplete-timeout = 15000, speech-complete-timeout = 1000, speech-start-timeout = 50, speech-language = es-MX}builtin:dtmf/digits\nbuiltin:speech/transcribe);

caller_input = session:getVariable('detect_speech_result');

It is working well with dtmf and speech. There is only one more thing I would like to do: when using dtmf I would like the caller to be able to press # to signal when the entry is completed and go to the next prompt immediately.  What is happening now is that # does not really do anything. After the user presses # it waits until the timeouts are reached (in fact I have to clean it up from the result after to remove the #).  If I add session:setVariable('playback_terminators','#') before  play_and_detect_speech, the result is that whatever was captured is overridden by the  playback_terminator.  In other words, the detect_speech_result variable does not come back with the digits entered by the customer but with (literally) "DIGITS: #" instead.

Do you have any tricks in your bag of tricks I can use?

I hope the above is clear.

Thanks!