Selected media is decoded in browser memory. Application and model files still download over the network.
Speech recognition in the browser: how local processing works
Speech recognition converts spoken audio into written text. In Whisper Web, the browser decodes the media, runs a Whisper model on the device and stores the finished transcript locally.
Choose a recording01
How local speech recognition works
The media path stays inside the browser from selection to editable transcript.
Decode the media
The browser reads a file, microphone recording or CORS-enabled direct media URL and converts it to audio samples.
Run Whisper locally
A Web Worker runs the selected Tiny, Base or Small model through WebAssembly or WebGPU.
Keep the transcript local
The editable text and timestamps stay in IndexedDB until you export or delete the record.
02
Local vs cloud speech recognition
The main difference is where the audio is processed and where the result is retained.
| Question | Whisper Web local mode | Typical cloud service |
|---|---|---|
| Where is audio processed? | On the current device | On a remote server |
| Is an upload required? | No media upload | Usually yes |
| What provides compute? | Browser WebAssembly or WebGPU | Provider infrastructure |
| What are the trade-offs? | Model download, device speed and browser limits | Network, account, pricing and provider policies |
03
Privacy and compatibility boundaries
Local processing removes the media-upload step, but it does not make the entire website offline.
The free local workflow rejects larger files and media longer than 20 minutes before transcription.
WebAssembly provides the compatible default. WebGPU depends on browser and device support.
04
Speech recognition FAQ
Does browser speech recognition upload my audio?
Not in Whisper Web local mode. The browser decodes the selected media and runs Whisper on the device, although it still downloads the application and model files.
Is WebGPU required for local speech recognition?
No. Whisper Web offers WebAssembly for broad compatibility and WebGPU as an optional faster backend on supported browsers and hardware.
What happens after the transcript is created?
The transcript is stored in the current browser using IndexedDB until the user exports or deletes it.
Transcribe on this device
Choose a recording