
Local Whisper in the Browser: Speech Recognition with WebGPU and WebCodecs
Quality speech recognition on the web still feels permanently chained to server GPUs and paid cloud APIs: try running a heavy neural net inside a browser tab, and it either freezes solid or devours all available RAM. The browser-whisper library flips that logic on its head.
The secret is an elegant pairing of WebGPU and WebCodecs. Instead of shipping multi-megabyte FFmpeg builds, audio is decoded by native browser codecs while matrix math runs directly on the user's GPU. Splitting the pipeline across two independent Web Workers keeps the UI responsive down to the millisecond while the model processes the stream.
The payoff is zero server infrastructure cost and complete privacy—audio never leaves the client device.






