A speech detector running in a worker (DECISIONS #167): Silero VAD (MIT, 0.64 MB, downloaded
once and cached). Measured: speech found 98–100 % with music or fan noise as loud as the voice
(judging by loudness found 0–14 %), music or noise alone never taken for speech; ~0.1 s per
22 s of sound. createCaptionsEndpoint uses one by default.
A speech detector running in a worker (DECISIONS #167): Silero VAD (MIT, 0.64 MB, downloaded once and cached). Measured: speech found 98–100 % with music or fan noise as loud as the voice (judging by loudness found 0–14 %), music or noise alone never taken for speech; ~0.1 s per 22 s of sound.
createCaptionsEndpointuses one by default.