❓ Questions / Help / Support

[eliran-fm]

max_speech_duration_s behavior

Hey,
In your documentation you specify the max_speech_duration_s parameter, and state that chunks longer than that will be attempted to be split at the last silence that lasts more than 100s, did you mean 100ms? Because in the code I can see the following:
min_silence_samples_at_max_speech = sampling_rate * 98 / 1000

Thanks

[snakers4]

min_silence_samples_at_max_speech = sampling_rate * 98 / 1000

Hi, this parameter is used here https://github.com/snakers4/silero-vad/blob/master/utils_vad.py#L317 for a different reason

[snakers4]

but also need some parameter to set the silence required for non-aggressive cut, which from the doc I understand is hardcoded to 100s (and I suspect you meant 100ms instead)

I cannot see such parameter in the code, can you please point to it?

[eliran-fm]

My understanding was that min_silence_samples_at_max_speech is exactly for that (the name also states that)

From the doc:
Chunks longer than max_speech_duration_s will be split at the timestamp of the last silence that lasts more than 100s (if any), to prevent agressive cutting.
Meaning there's the max_speech_duration_s which determines the maximum output chunk duration, and some variable (which I presume is min_silence_samples_at_max_speech) that determines the aforementioned "last silence" duration, and is currently fixed to 98 milliseconds

[snakers4]

min_silence_samples_at_max_speech = sampling_rate * 98 / 1000
which from the doc I understand is hardcoded to 100s

This is about 100ms, not 100 seconds

[eliran-fm]

Ok I see, so maybe it's worth updating the doc
Thank you for clarifying @snakers4 !

[dayfinihzy]

Where can I find the doc?