snakers4 / silero-vad

Silero VAD: pre-trained enterprise-grade Voice Activity Detector
MIT License
4.43k stars 432 forks source link

Speech Probability Threshold below 0.15 causes negative Inactivity Threshold #580

Closed Yohrog closed 1 day ago

Yohrog commented 2 days ago

https://github.com/snakers4/silero-vad/blob/cff5eb2980d64e80dca840277528f1a17bb6193c/src/silero_vad/utils_vad.py#L311

Since neg_threshold will be set with a fixed value of threshold - 0.15 when not given, neg_threshold will be negative for threshold values less than 0.15.

It'd be good to either set it relative with something like threshold * 0.7 or at least give a warning to the user.

snakers4 commented 2 days ago

Good catch.

adamnsandle commented 1 day ago

Fixed in https://github.com/snakers4/silero-vad/pull/581/commits/eaf633ec9d6594544707c6ea21cedcf9558cb1a3