README.md

metadata

license: mit
pipeline_tag: voice-activity-detection
tags:
  - onnx
  - audio
  - silero-vad
  - transformers.js
library_name: transformers.js

BricksDisplay/silero-vad-6.2

Transformers.js-compatible Silero VAD v6.2 packaged for ONNX inference.

Files

• onnx/model.onnx — fp32
• onnx/model_fp16.onnx — fp16
• onnx/model_int8.onnx — dynamic int8
• onnx/model_uint8.onnx — dynamic uint8
• onnx/model_quantized.onnx — alias of uint8 for Transformers.js dtype: "q8"

Source

• Upstream assets: snakers4/silero-vad tag v6.2
• Reference packaging layout: BricksDisplay/silero-vad

Transformers.js

import { AutoModel } from '@huggingface/transformers';
const model = await AutoModel.from_pretrained('BricksDisplay/silero-vad-6.2');

Inputs expected by the ONNX session:

• input: float32 PCM chunk, shape [1, num_samples]
• state: float32 recurrent state, shape [2, 1, 128]
• sr: int64 scalar sample rate (8000 or 16000)