Offline transcription
Turn recordings, voice memos and videos into text with the Whisper speech model — running on your own device, so interviews, meetings and medical or legal audio never leave it.
The speech model (about 40 MB) downloads once from this site the first time you transcribe.
The Whisper model runs on your device and is loaded from this site. Your audio is never uploaded — open the Privacy check at the bottom of the page to watch.
How it works
This tool runs Whisper (the open-source speech model from OpenAI, “tiny” size, 39 million parameters) with Transformers.js and ONNX Runtime compiled to WebAssembly. The model files and the runtime are served by this site — the page isn’t even allowed to contact Hugging Face or any other server — and your audio is decoded and transcribed inside your browser.
What to expect
- Accuracy: good for clear English; weaker for heavy accents, crosstalk, background noise or less common languages. The tiny model trades accuracy for speed so it can run on ordinary laptops and phones.
- Speed: on a recent laptop, a minute of audio usually takes 10–40 seconds. Phones are slower.
- First use downloads about 40 MB of model files, which your browser then caches.
Why offline matters
Cloud transcription services receive your recordings. For interviews, therapy notes, medical appointments, legal calls or anything confidential, running the model on your own device keeps the audio private by design. Turn on airplane mode after the model has loaded and it still works.
Need to pull audio from a video first? The video to MP3 tool also runs locally. Count or tidy the transcript with the word counter.
Frequently asked questions
Is my audio uploaded?
No. Audio is decoded and transcribed in your browser. The page’s security policy only allows it to load files from this site, and the Privacy check lists every request, including those made by the transcription engine.
Which languages are supported?
Whisper recognises about 99 languages. English works best with this small model; French, Spanish and other widely spoken languages are usable for clear audio. You can also translate speech into English text.
Can I make subtitles?
Yes. Download the transcript as SRT or VTT for video players, YouTube or editing software.