Skip to content
Apoorv Mittal
Back to experiments
Lab · on-device speech recognition

Local Whisper

Speech recognition without the cloud: Whisper runs inside your browser on WebGPU (or WASM as fallback). The model downloads once from Hugging Face's CDN; after that, every byte of audio is processed on your machine. The counterpoint to the Web Speech API, which Chrome and Edge route through vendor servers.

This downloads the Whisper base model (~80 MB) from Hugging Face's CDN into your browser's cache. Nothing starts until you click.

The model is fetched from Hugging Face's CDN and its runtime from jsDelivr; your audio and transcript never leave this browser tab. Nothing is sent to or stored by this site.