Audio to Text (Transcribe)
What the transcriber does
Drop in a recording — an interview, voice memo, lecture, meeting, or podcast — and this tool writes out what was said as text you can copy or save. A compact speech-recognition AI model (OpenAI’s Whisper, in its browser build) runs on your own device to listen to the audio and produce a transcript, with no typing and no service to sign up for.
Everything stays on your device
Most transcription services upload your audio to their servers. This one does not: the model runs in your browser using WebAssembly, so your file is decoded and transcribed locally and never leaves your computer. That makes it a good fit for sensitive recordings — private interviews, medical or legal notes, or anything you would rather not hand to a third party.
The one-time model download
The first time you transcribe, the browser fetches the Whisper model (about 150 MB) and caches it. After that, transcriptions start without re-downloading and even work offline. Only the model is downloaded — your audio is never sent anywhere.
What to expect on accuracy
This tool uses a small, fast English model, so it is excellent for notes, drafts, and making a recording searchable, but it can slip on names, technical jargon, or noisy audio. For an important transcript, read it through and fix the occasional word. It reads common audio (MP3, WAV, M4A, OGG) and video (MP4, MOV, WebM) files up to 100 MB.
Frequently asked questions
Is my audio uploaded to a server?
No. The speech-recognition AI runs entirely in your browser using WebAssembly, so your audio (or video) file is transcribed on your own device and is never uploaded.
What is downloaded on the first run?
Only the AI model — a Whisper speech-recognition model (~150 MB) — is fetched once and then cached by your browser, so later transcriptions work without re-downloading and even offline. Your file itself is never sent anywhere.
What languages does it handle?
This tool uses an English speech model, so it works best on clear English speech. Very noisy recordings or heavy background music can reduce accuracy.
What file types can I transcribe?
Common audio (MP3, WAV, M4A, OGG) and video (MP4, MOV, WebM) files up to 100 MB. The audio is decoded locally before transcription.
How accurate is it?
It uses a small, fast model, so it is great for notes, drafts, and searchable transcripts, but it can make mistakes on names, jargon, or unclear audio. Always proofread an important transcript.
Does it work on phones?
It works best on a laptop or desktop. Because the AI model runs in your browser and needs a fair amount of memory (~150 MB model plus working memory), phone browsers can run out of memory mid-transcription and reload the tab. If that happens, try the same file on a computer.