| license: mit | |
| tags: [voice-activity-detection, silero, onnx, betterflow] | |
| # betterflow — Silero VAD | |
| `silero_vad.onnx` as distributed in the [k2-fsa/sherpa-onnx](https://github.com/k2-fsa/sherpa-onnx) | |
| `asr-models` release. Mirrored here **only** so betterflow can pin an immutable revision and verify a | |
| sha256 at install time; the weights are unmodified. | |
| Used to cut a dictated utterance into VAD-endpointed phrases so each decodes independently, which | |
| lets the first text reach the user without waiting for the whole decode. | |
| - ONNX input `x` is fixed at shape **[1, 512]** — 512 samples at 16 kHz, per call. Not configurable. | |
| - Self-describes as silero-vad **v4** exported by k2-fsa, 16 kHz branch only. | |
| - sha256: `9e2449e1087496d8d4caba907f23e0bd3f78d91fa552479bb9c23ac09cbb1fd6` | |
| - Upstream: https://github.com/snakers4/silero-vad (MIT) | |