Voice Activity Detection

Detect speech regions in an audio file. Two output modes:

1. Sidecar mode (default) — write a JSON + CSV of [start, end] speech segments next to the input.
2. Split mode — cut the input into per-utterance files in a vad-segments/ folder.

Arguments ($ARGUMENTS)

• <input> (required) — path to an audio file (any format ffmpeg handles).
• --mode=<sidecar|split> (optional, default sidecar).
• --out=<dir> (optional) — output directory. Defaults to the input's directory for sidecar mode, <input-dir>/vad-segments/ for split mode.
• --min-speech=<seconds> (optional, default 0.3) — minimum segment length to keep.
• --min-silence=<seconds> (optional, default 0.5) — minimum silence to treat as a break.
• --threshold=<dB> (optional, default -40) — peak threshold below which counts as silence (used when falling back to ffmpeg).
• --engine=<auto|silero|webrtc|ffmpeg> (optional, default auto).

Engine selection