Silero VAD v5 (ONNX) — Solus mirror

Byte-identical mirror of the ONNX export from onnx-community/silero-vad, served here so Solus installs never depend on a third-party repository staying available.

Used by Solus for dictation voice-activity detection, replacing an RMS loudness threshold.

Contract (verified by probing the graph)


`input`	f32 `[batch, 64 + 512]` — 64 samples of previous audio, then the frame
`state`	f32 `[2, batch, 128]` — recurrent state, threaded call to call
`sr`	int64 scalar — 16000
`output`	f32 `[batch, 1]` — speech probability
`stateN`	f32 — the next `state`

Frames are exactly 512 samples at 16 kHz (32 ms). The graph accepts 480 and returns a plausible number, so a wrong frame size fails silently rather than loudly.

Licence: MIT, as upstream.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

Voice Activity Detection

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Fazmin/solus_v1_silero-vad-v5

Base model

onnx-community/silero-vad

Quantized

(9)

this model