About 100 languages
Whisper recognizes 99 languages, from English, Spanish and Chinese to Swahili and Welsh. Accuracy is best for widely spoken languages and improves with the larger models.
Drop in a recording and get an editable transcript with timestamps. Whisper speech recognition runs right here in your browser — nothing is uploaded.
Three steps. Your recording never leaves this page.
Pick an audio file — MP3, M4A, WAV, FLAC, OGG — or a video. It's opened by your browser, not sent to a server.
Leave the language on Auto-detect or choose it. The first time, your browser downloads the speech model and keeps it for next time.
Lines appear as they're transcribed. Fix any word, click a timestamp to hear that part, then download TXT, SRT or VTT.
| Price | Free, with no limit on the number of files |
|---|---|
| Sign-up | Not needed |
| Audio in | MP3, M4A, AAC, WAV, FLAC, OGG, Opus — and MP4, MOV, MKV, WebM video |
| Text out | Editable lines with timestamps; download TXT, SRT or VTT, or copy the text |
| Languages | 99, detected automatically or chosen by you |
| Where it runs | In your browser, on your device — the file is not uploaded |
| Length | No fixed limit on a computer; on a phone, about 10 minutes works best |
| Speaker labels | Coming with paid cloud transcription |
Interviews, lectures and screen recordings often live in video files. Video to text takes MP4, MOV, MKV and WebM, pulls out the sound in your browser and transcribes it the same way.
The subtitle converter switches between SRT, VTT and plain text, and pulls text subtitles out of MKV files.
Whisper recognizes 99 languages, from English, Spanish and Chinese to Swahili and Welsh. Accuracy is best for widely spoken languages and improves with the larger models.
Each line of the transcript has a start and end time, so you can jump to it, check it, and export it as subtitles.
Click any line to correct names and terms. Play the recording from any timestamp to check what was said. Your edits go into every download.
Download plain text for notes and articles, or SRT and VTT subtitle files with timings for video editors and players. All formats are free.
The speech model runs on your own device. Your audio isn't uploaded, stored or used to train anything — there is no server that could. The only downloads are the model files, the first time you use each model.
Journalists and researchers turn recorded interviews into text they can quote, search and code — without sending sensitive conversations to a third party.
Students transcribe lecture recordings to study from and search, and teachers make text versions of their lessons.
Get a written record of a call you recorded, then copy the parts you need into notes or minutes.
Podcasters make episode transcripts for show notes and accessibility, and SRT files for video versions.
Turn ideas you dictated on your phone into text you can edit. iPhone voice memos are M4A files and work directly.
Yes. Transcription runs on your own computer or phone, so it costs us nothing to run and there's no limit on how many files you transcribe. All export formats are included.
No. Open the page, choose a file, and transcribe. Nothing to sign up for.
The speech recognition model (OpenAI's Whisper, in a browser-ready version) runs in your browser instead of on a server. Your browser downloads it the first time — Fast is about 120 MB on a computer and about 40 MB on a phone; Best is about 560 MB — and keeps it, so next time it starts right away.
Usually, for shorter recordings. Phones use the Fast model and are slower than computers; keep the tab open while it works. For long recordings, a desktop computer with Chrome or Edge is much faster.
MP3, M4A, AAC, WAV, FLAC, OGG and Opus audio, and MP4, MOV, MKV and WebM video. Which ones decode depends on your browser; desktop Chrome and Edge handle all of them, and Dolby (AC-3) audio tracks may not decode. There is no fixed length limit — the audio is processed in pieces, so long files work on a computer; they just take longer.
The 99 languages Whisper was trained on. Auto-detect picks the language from the recording; if it guesses wrong, choose the language yourself.
On a computer with a recent GPU, transcription is much faster than real time: in our tests an hour-long recording took about 3 minutes with the Best model on a recent Mac laptop. The first time, your browser also downloads the model, which takes from under a minute to a few minutes depending on your connection and the model. Phones and computers without WebGPU are slower.
It depends on the model and the recording. Clear speech with one person close to the microphone comes out very well; background noise, overlapping speakers, heavy accents and specialist terms cause more mistakes. Fast is the default because it starts quickly; if words or names come out wrong, run the file again with Standard. Always read the transcript before you rely on it.
Speaker labels are coming with paid cloud transcription. The free in-browser transcriber gives you timestamped lines, and you can add speaker names as you edit before you download.
Use Copy text and paste into Word or Google Docs, or download TXT, which opens in any editor. For subtitles with timings, download SRT or VTT.
No. The file is read by your browser and transcribed on your device. It isn't sent to us or anyone else, and nothing is used for training. See the privacy page for the details.
No. Please upload a file you have the rights to use, like your own recording.
Yes. When the transcript is ready, choose SRT (or VTT). To change subtitle formats later, use the subtitle converter.
Free, no sign-up, and it never leaves your device.
Choose a file