use case · multi-model

Multi-model: VAD → Whisper, transcribe speech only

This is why VAD earns its keep. Whisper is expensive; silence is free to skip. Silero finds the speech segments first, then Whisper transcribes only those — so the transcriber never burns time on dead air. Two models, both on-device: a tiny recurrent VAD and a Transformer ASR, composed.

Two models, both local

Silero VAD (~2 MB, onnxruntime-web) segments; Whisper-base (transformers.js) transcribes. Both auto-initialise from cache on a return visit.

1 · Silero VAD

2 · Whisper (ASR)

Drop an audio file, or click to choose

The bundled Speech clip alone (not the other samples) is a JFK inaugural excerpt (20 January 1961), spoken by John F. Kennedy — public domain, U.S. government work · JFK audio source and attribution record.

Timeline (green = speech VAD sends to Whisper):

← Wild · Back to Silero VAD