Expand description
Voice activity detection: a multi-feature, epistemic proposal stack (energy above an adaptive noise floor + spectral flatness + ZCR), with hysteresis/hangover segmentation. Silence is never coerced into speech.
Re-exports only (AU-VAD). Zero-heap hot path (caller-buffered / stack scratch).
Re-exports§
pub use frame_vad::frame_is_voiced;pub use frame_vad::frame_voicing_score;pub use frame_vad::MAX_VAD_FFT;pub use noise_estimate::noise_floor_min_stat;pub use noise_estimate::NOISE_BIAS;pub use segmenter::segment_voiced;
Modules§
- frame_
vad - Per-frame voice-activity decision from a fusion of three cheap, orthogonal cues:
- noise_
estimate - Adaptive noise-floor estimation by minimum statistics.
- segmenter
- Segment a mono signal into voiced regions with hysteresis and hangover.