Clarification on how silence detection works (silence_short, silence_medium, silence_long, silence)

[Maciej Zasadzki ADVICOS]

Hi,

I’d like to better understand how silence detection works in VoiceGuide.

In the module shown in the logs I’ve sent (attached), the parameter silence_long was used. However, during the call, only silence_short and silence_medium events appeared in the traces — there was no silence_long event.

Could you please explain how the silence detection mechanism operates internally?
Specifically:
– What is the exact relationship between silence_short, silence_medium, silence_long, and the generic silence event?
– Does the system evaluate them sequentially (e.g., short → medium → long), or are they independent detectors active in parallel?
– When I define silence_long in a Play module that performs speech recognition, does it mean only this one will be reported, or are the others still active in the background?
– Is there perhaps a diagram, timing chart, or logic flow that shows how these silence levels are triggered over time during an STT session?

Understanding this would help me set the right expectations and configuration for longer versus shorter user responses.

Thanks in advance for your explanation!

[SupportTeam]

A silence_short event will be reported if a silence (no speech) of length stt_silence_length_short is detected during the first stt_silence_window_short while doing STT.

A silence_medium event will be reported if a silence (no speech) of length stt_silence_length_medium is detected during the first stt_silence_window_medium while doing STT.

A silence_long event will be reported if a silence (no speech) of length stt_silence_length_long is detected at any time during STT.

A silence event is also raised whenever any of the above silence_* events occur.

The above lets you use the simple silence detection approach to detect end of callers response faster when caller gives short responses. Like Yes/No or a short statement. 

 

Note that these days there are much better methods available for detecting when the caller has concluded speaking. You would need to contact sales@voiceguide.com to discuss.