How to transcribe audio and video
Drop an audio or video file (or record one), press Transcribe, and the text appears line by line with the time each line was said. Correct any word by clicking it, then save the transcript as TXT, or as SRT or VTT subtitles.
Add a recording
A voice memo, an interview, a lecture, a meeting, a podcast or a video. Only the sound is used, and it is read straight from the file, so long videos work too.
Choose the model and language
Fast (Whisper base, a 94 MB download) suits clear speech. Better (Whisper small, 252 MB) makes fewer mistakes, especially in other languages, and takes about four times as long. Leave the language on automatic, or pick it when the recording starts with music or silence.
Transcribe, correct, save
The first time, the model downloads once and stays on your device. Lines appear as each half minute is done. Click a time to hear that moment, click the text to fix it, then save.
Subtitles for a video
Save SRT for most video editors and YouTube, or VTT for web players. Each subtitle keeps to two lines of at most 42 characters, the usual limit for readable subtitles; longer sentences are split into several cues with the time shared out by length. Corrections you make in the transcript go into the saved file.
Languages and translation
Whisper understands 99 languages and works out which one is spoken from the first half minute. It is strongest in English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Polish, Japanese, Chinese and Korean, and weaker in languages with less training audio. Turn on "Translate into English" to get an English transcript of speech in any of them.
How accurate it is
Clear speech with one person at a time comes out nearly word for word, with punctuation. Background music, noise, people talking over each other, strong accents and rare names cause mistakes. Long silences are skipped, because speech models tend to invent words in silence. The Better model is the one to use for anything you will publish.
Where does the recording go?
It stays on your device. The Whisper model downloads from this site once and runs right in your browser. A microphone recording is kept only until you close the page.
Models and licences
| Fast | Whisper base by OpenAI, MIT licence, in onnx-community's 8-bit export (77 MB) |
|---|---|
| Better | Whisper small by OpenAI, MIT licence, in onnx-community's 8-bit export (250 MB) |
| Reading the file | mediabunny (MPL-2.0) |
| Runtime | onnxruntime-web (MIT) |
Specs and browser support
| Input | MP3, WAV, M4A, AAC, OGG, Opus, FLAC; the sound of MP4, MOV, WebM and MKV videos; a microphone recording |
|---|---|
| Output | Plain text, SRT and WebVTT, UTF-8 |
| Speed | On a recent laptop, Fast transcribes an hour in roughly five to ten minutes and Better in roughly twenty to forty, depending on the processor |
| Length | No limit set; about an hour needs 250 MB of memory for the sound |
| Browsers | Current Chrome, Edge, Firefox and Safari on a computer |
| Price | Free, no watermark, no account |
Questions
No. The Whisper model is downloaded to your browser and runs there. Your audio and video stay on your device.
Clear speech in English and other widely spoken languages comes out nearly word for word. Noise, music, overlapping voices and rare names cause mistakes; the Better model makes fewer. Each line of the transcript can be corrected before saving.
Yes. Drop the video, transcribe it and save SRT or VTT. The cues keep to two lines of 42 characters, which video players and YouTube accept.
Not yet. Whisper writes what is said and when, but not who said it.