Skip to content

Distilled English-only Whisper Small — smallest Distil-Whisper checkpoint at 166M params with 4 decoder layers. Within 3% WER of Whisper Large v2 and 5.6x faster (relative latency vs Large v2), per the Distil-Whisper project's own evals. English-only alternative that trades multilingual coverage for accuracy on the language most demos use.

WebGPU
WASM
Streaming
MIT

Try It

DistilWhisper Small (English)
Not loaded(172.0 MB)

Download the model first to start recording

Technical Details

Architecture

Whisper (distilled encoder-decoder)

Parameters

166M

Model Size

172 MB

Languages

en

Links