mediainbox/silero-vad-onnx

Silero VAD

Torch-free ONNX Runtime weights for Silero VAD 6.2.1.

Models

• silero_vad.onnx: official model for streaming at 8 or 16 kHz.
• silero_vad_16k_sequence.onnx: faster offline model for 16 kHz audio.
• silero_vad_8k_sequence.onnx: faster offline model for 8 kHz audio.

Usage

from silero_onnx_vad import SileroVAD, get_speech_timestamps

model = SileroVAD.from_pretrained(
    "mediainbox/silero-vad-onnx",
    filename="silero_vad_16k_sequence.onnx",
)

segments = get_speech_timestamps(
    audio,
    model,
    sampling_rate=16_000,
)

Use the sequence models for long offline recordings and the official model for live streaming.