❓ Can window_size_samples be selected as 160 (10ms)?

[jifashen]

❓ Questions and Help

For 16000Hz audio, window_size_samples can choose 512 (32ms) 1024 (64ms) 1536 (96ms). Can window_size_samples be selected as 160 (10ms)? In addition, what do the parameters threshold, min_silence_samples_at_max_speech, min_speech_samples, max_speech_samples, speech_pad_samples mean, and what impact do they have on the vad results?

[snakers4]

Can window_size_samples be selected as 160 (10ms)?

No, because we do not have data annotation with such granularity.

what do the parameters threshold, min_silence_samples_at_max_speech, min_speech_samples, max_speech_samples, speech_pad_samples mean

threshold - it is the main activation parameter, probability threshold when the VAD activates
min_speech_samples - the minimal number of audio samples that can constitute speech, this is to suppress spurious activations
max_speech_samples - the same but for maximum speech length

min_silence_samples_at_max_speech, speech_pad_samples - these are a but technical, it is better to look them up in the utils code

and what impact do they have on the vad results?

They mostly correct noise and make outputs more robust and suppress spurious activations.