jambonz / freeswitch-modules Goto Github PK

we noticed that for small intent like Yes/no yeah etc Microsoft is sending no_speech_detected due to which jambonz is not sending any received transcript to bot due gather is getting timeout

2024-02-07 01:22:39.525528 84.63% [DEBUG] azure_transcribe_glue.cpp:177 GStreamer onRecognitionEvent reason 0 results: {"Id":"b5d353ed9b754f8aa5ef4936dbaa4c29","RecognitionStatus":"InitialSilenceTimeout","Offset":0,"Duration":158400000,"Channel":0,"DisplayText":"","NBest":[{"Confidence":0.0,"Lexical":"","ITN":"","MaskedITN":"","Display":""},{"Confidence":0.050563347,"Lexical":"no","ITN":"no","MaskedITN":"no","Display":"no","Words":[{"Word":"no","Offset":157300000,"Duration":800000,"Confidence":0.050563347}]},{"Confidence":0.05015686,"Lexical":"yes","ITN":"yes","MaskedITN":"yes","Display":"yes","Words":[{"Word":"yes","Offset":157300000,"Duration":1100000,"Confidence":0.05015686}]},{"Confidence":0.05075142,"Lexical":"please","ITN":"please","MaskedITN":"please","Display":"please","Words":[{"Word":"please","Offset":156100000,"Duration":2300000,"Confidence":0.05075142}]},{"Confidence":0.050710723,"Lexical":"to","ITN":"to","MaskedITN":"to","Display":"to","Words":[{"Word":"to","Offset":157300000,"Duration":800000,"Confidence":0.050710723}]}]},

2024-02-07 01:22:39.525528 84.63% [INFO] mod_azure_transcribe.c:20 responseHandler event azure_transcribe::no_speech_detected, body {"Id":"b5d353ed9b754f8aa5ef4936dbaa4c29","RecognitionStatus":"InitialSilenceTimeout","Offset":0,"Duration":158400000,"Channel":0,"DisplayText":"","NBest":[{"Confidence":0.0,"Lexical":"","ITN":"","MaskedITN":"","Display":""},{"Confidence":0.050563347,"Lexical":"no","ITN":"no","MaskedITN":"no","Display":"no","Words":[{"Word":"no","Offset":157300000,"Duration":800000,"Confidence":0.050563347}]},{"Confidence":0.05015686,"Lexical":"yes","ITN":"yes","MaskedITN":"yes","Display":"yes","Words":[{"Word":"yes","Offset":157300000,"Duration":1100000,"Confidence":0.05015686}]},{"Confidence":0.05075142,"Lexical":"please","ITN":"please","MaskedITN":"please","Display":"please","Words":[{"Word":"please","Offset":156100000,"Duration":2300000,"Confidence":0.05075142}]},{"Confidence":0.050710723,"Lexical":"to","ITN":"to","MaskedITN":"to","Display":"to","Words":[{"Word":"to","Offset":157300000,"Duration":800000,"Confidence":0.050710723}]}]}.

2024-02-07 01:22:39.525528 84.63% [DEBUG] mod_azure_transcribe.c:26 responseHandler returning event azure_transcribe::no_speech_detected, from finished recognition session

Connect to a call that is using ElevenLabs TTS and OPUS codec and you'll notice the tts is garbled and sounds like chipmunks.

when we set en-IN as the language for google ASR, it does not give transcriptions. en-US works. Our assumption is enhanced model is being set to true by default in transcribe. This could be tested through transcribe and setting language to en-IN