Pluggable per-channel voice-activity detector. The default EnergyVad is energy-based
with an adaptive noise-floor threshold; callers can drop in a DNN-backed detector
(e.g. Silero via @ricky0123/vad-web) for noisier environments.
A separate VadDetector instance is held per channel; do not assume cross-channel
state. Frames are fixed-size at the transcriber's target sample rate.
Pluggable per-channel voice-activity detector. The default
EnergyVadis energy-based with an adaptive noise-floor threshold; callers can drop in a DNN-backed detector (e.g. Silero via@ricky0123/vad-web) for noisier environments.A separate
VadDetectorinstance is held per channel; do not assume cross-channel state. Frames are fixed-size at the transcriber's target sample rate.