Pick an audio or video file and OpenAI's Whisper turns it into text inside this browser tab, on your own computer. Nothing gets uploaded. It's free, there's no account, and there's no minute limit.
Checking what your browser supports...
Rather talk than type? StarWhisper puts the same Whisper in every text box on your PC or Mac. Press a hotkey, speak, and the words appear in your email, doc or chat. Free for 2,000 words a week, no account needed.
Try it freeMost free transcription sites upload your recording to their servers first. This page doesn't. The audio is decoded by your browser and handed to Whisper running in the same tab. You can check it yourself: open your browser's developer tools, go to the Network tab, and transcribe something. You'll see the model files coming in from huggingface.co the first time, and no upload going out.
That makes it a reasonable choice for interviews, client calls, medical or legal notes, and anything else you'd rather not hand to a third party.
Two things do leave the page, and neither contains your file or your words. If you accept cookies, our usual site analytics load as they do on every page. And when a transcription finishes, the page sends one anonymous count to our server with the model, CPU or GPU, and a rough length such as "1 to 5 minutes", without a cookie. That's how we know whether anyone uses this. The code is open source on GitHub, so you can read exactly what it does.
| Model | Download on CPU / GPU | Good for |
|---|---|---|
| Tiny | 41 MB / 120 MB | Quick drafts, clear English speech, older computers and phones |
| Base | 77 MB / 206 MB | Most recordings with one or two clear speakers |
| Small | 249 MB / 586 MB | Accents, other languages, noisy rooms, when accuracy matters more than speed |
The GPU versions are bigger because they store the weights in a format graphics cards run well. All three models are multilingual and cover the 99 languages Whisper was trained on. If the result has a lot of mistakes, try the next model up before anything else.
Anything your browser can play should work: MP3, WAV, M4A, AAC, OGG, Opus, FLAC and WEBM audio, plus the audio track of MP4 and WEBM videos. Chrome and Edge read the most formats. If a file won't open, convert it to MP3 or WAV first.
There's no minute limit, but your computer does the work, so it sets the pace. A graphics card through WebGPU is much faster than the processor alone, and on a processor a long recording can take longer than the recording itself. Keep the tab open until it finishes. Files up to 1 GB load on a computer and up to 200 MB on a phone. For a long video, save just the audio as MP3 first; it's far smaller and decodes faster.
This page is for files you already have. If you want to dictate, the StarWhisper desktop app is the same Whisper running as a native app on Windows and Apple Silicon Macs. You press a hotkey, talk, and the text appears in whatever app you're typing in: email, Word, Slack, your code editor. It uses CUDA, Vulkan or Metal for speed and works offline once set up. The free plan covers 2,000 words of dictation a week and needs no account. Pro adds file transcription inside the app, without a browser's memory limits, plus the Medium and Large models.
Download StarWhisper or read how it compares to running whisper.cpp yourself.
Yes. Your computer does the work, so there's nothing for us to pay per minute and no reason to cap you. There's no account and no watermark on the text.
No. The file is read and transcribed inside your browser. The only things downloaded are the page, the transcription library from jsDelivr and the model files from Hugging Face.
Yes. On automatic, Whisper listens to the first 30 seconds and picks the most likely language. If a recording starts with music or silence, choose the language yourself for better results.
Whisper can translate speech from other languages into English text. Tick "Translate the result into English" before you pick the file. It can't translate into other languages.
No. Whisper writes down what was said, not who said it.
Usually because the browser has no WebGPU, so Whisper runs on the processor. The note under the file box tells you which one you're on. Chrome or Edge on a recent Windows or Mac computer is the fastest setup, and the Tiny model is the quickest of the three.
It can, on recent phones with a current browser, but phones have far less memory, so the page offers Tiny and Base only there. Shorter recordings work best.