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 an audio or video fileHow does local speech recognition work?
You choose the media, the browser runs Whisper, and you review the transcript on the same device.
Decode the media
The browser reads a file, microphone recording or publicly accessible 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.
How does local speech recognition compare with cloud processing?
Start with where the audio goes, who provides the compute and how you want to share the result.
| 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 |
What are the privacy and compatibility limits?
Local processing removes the media-upload step. The page and model still need the network, and your browser and device set practical limits.
- No media upload
Selected media is decoded in browser memory. Application and model files still download over the network.
- Choose the tool that fits the recording
The standard tool accepts up to 300 MB and 20 minutes. For a local file up to 1 GB and 1 hour, use the large-file tool.
- WebGPU is optional
WebAssembly provides the compatible default. WebGPU depends on browser and device support.
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.