tonythethompson/Silero-VAD-INT8-ONNX

Silero VAD — INT8 ONNX

INT8-quantized ONNX export of Silero VAD, an open-weight voice activity detection model. The immediate upstream source is onnx-community/silero-vad; the original model is from the Silero team (snakers4/silero-vad on GitHub). This is not a newly trained model.

Source

Field	Value
Upstream model	onnx-community/silero-vad
Upstream source revision	e71cae966052b992a7eca6b17738916ce0eca4ec
Export tool/script	Mirrored from onnx-community (no re-export)
Quantization recipe	Pre-quantized INT8 ONNX mirrored from onnx/model_int8.onnx
Original model	snakers4/silero-vad (GitHub)

Files

File	Size	Description
onnx/model_int8.onnx	~624 KB	INT8 ONNX model

Intended Use

Voice activity detection for local inference via ONNX Runtime. The INT8 variant is the CPU/low-resource build: smaller and faster on CPU than the FP16 variant. The FP16 variant is in tonythethompson/silero-vad-fp16-onnx. Silero VAD is language-agnostic and works across a wide range of languages.

Runtime Notes

• Designed for ONNX Runtime compatible runtimes.
• Expected input: 16 kHz mono audio (Silero VAD requirement).
• INT8 is broadly compatible across CPU hardware (x86-64, ARM). Validate on the target execution provider before production use.

Precision and Packaging

Export tooling, precision, and quantization are recorded in the Source table above. This packaging mirror does not publish independent parity benchmarks; validate on your target execution provider before production use.

Limitations

• Requires 16 kHz mono audio input; other sample rates must be resampled.
• INT8 may reduce detection sensitivity at the margins (very quiet or very short speech).
• No repository-specific VAD quality benchmark is documented here.

License

MIT — inherited from onnx-community/silero-vad and the original Silero VAD model. This packaging repo adds no new license terms.