True once the hysteresis has flipped the state to speech.
Feed audio into the detector. The configured callbacks fire on speech↔silence transitions after hysteresis stabilises.
StaticresetReset the per-process warning memoisation. Tests call this in
beforeEach so the one-shot-warning assertion is reproducible
across runs.
Incremental VAD over PCM16 @ 24 kHz mono audio.
Feed chunks via process; the configured
onSpeechStart/onSpeechEndcallbacks fire when the speech/silence transitions stabilise. The detector is purely sample-driven — no async, no timers — so a single instance is safe to share across an entire scenario as long as one adapter feeds it.