Transcribe Audio in Your Browser — Nothing Uploaded

No. The file is decoded in your browser and the model runs on your device. This is the whole reason the tool exists — the recordings people most need transcribed are usually the ones they are not allowed to upload anywhere.

With WebGPU, roughly real-time or better on the base model — a ten-minute recording in a few minutes. On the WebAssembly fallback expect several times the audio duration. The first run also includes the model download.

At the large-v3-turbo setting this runs the same family of model most hosted services do, so on clear speech the gap largely closes. The hosted services still win on diarisation, on very noisy audio, and on turnaround for long files — but the model here is no longer the small end of the family. Move up a size before concluding the audio is the problem.

No. Whisper transcribes speech but does not do diarisation, so a two-person interview comes back as one continuous transcript. The segment timings make it straightforward to label afterwards.