Drop in a recording and get the words out of it. Whisper runs inside this tab, so the file never goes anywhere — not to a server, not to us.
MP3, WAV, M4A, OGG, FLAC, MP4, MOV, WebM — whatever your browser can play
The file you pick is decoded by the browser itself and handed to Whisper, OpenAI's speech recognition model, compiled to run as WebAssembly or on your graphics card. All of that happens on your machine. There is no upload step, which is why there is no per-minute price and no cap on how long your recording can be.
The first run downloads the model — anywhere from 40MB to 800MB depending on which one you pick. Your browser caches it, so every run after that is instant and works with the network off entirely. Bigger models hear better, especially with accents, background noise or technical vocabulary. Start small and move up if the result disappoints.
Speaking live works differently to a file. The page listens for you to start and stop talking, then runs each sentence through the model as soon as you pause — so text lands a second or so behind your voice, in bursts rather than word by word. Tiny or Base keep up comfortably; the larger models fall behind. Whatever you say is also recorded, so when you stop you get a playable file and the same editing and export options as any other transcript.
Timestamps are approximate. Whisper places them per phrase, not per word, and it drifts on long silences or overlapping speakers. Fine for subtitles and for finding your place in a recording. If you need frame-accurate captions, expect to nudge them.
Anything you type into the transcript stays in the page and disappears when you close the tab. Export before you go.
Every tool here is free and always will be. If one saved you a subscription, a coffee is a nice way to say so.
Buy me a coffee