Whisper online, running on your own computer
Transcribe with OpenAI's open-source Whisper models without installing anything. Whisper runs inside this browser tab on your own computer, so there is no upload, no queue and no length limit.
- Never uploaded
- No sign-up
- No daily limits
- Free
Keep this tab open and in front. Browsers slow down background tabs.
Click any line to fix a word. Downloads include your edits.
Quick answer: To transcribe with Whisper in your browser, drop the file into the tool at the top of this page and choose the spoken language. iLoveAudioVideo turns the speech into text on your own device, so the recording is never uploaded. It is free with no sign-up, and you can copy the text or download it as TXT, SRT or VTT.
Updated
How it works
- Pick the spoken languageEnglish is set. Choose another of 16 languages if needed.
- Drop the recordingMP3, M4A, WAV, MP4, MOV, OGG, OPUS or WEBM, of any length.
- Copy or downloadFix any word by clicking it, then copy the text or save TXT, SRT or VTT.
Which Whisper model you get
On your first visit the page times a short test on your computer (10–20 seconds) and picks the most accurate Whisper model that still runs at least 2.5 times faster than the recording plays. You can choose another one under Quality.
| Quality | Whisper model | One-time download |
|---|---|---|
| Fast | tiny | about 44 MB |
| Balanced | base | about 80 MB |
| High | small | about 250 MB |
| Best | large-v3-turbo (needs a capable graphics chip) | about 570 MB |
The model is stored in your browser after the first download, so later visits start quickly.
Why run Whisper in the browser
Hosted Whisper demos make you upload the recording and often wait in a queue. Running Whisper yourself usually means installing Python, the command line and the models. Here your own computer's processor, and its graphics chip when it has a suitable one, does the work in the page, and the recording never leaves your computer.
How accurate and how fast
On clear English speech we measured these word error rates: Fast 5–10%, Balanced 4–5%, High about 3–4%. Background noise, music and people talking over each other add errors.
A 60-minute recording took 15 minutes on a 4-core Windows computer without a graphics card, 9 minutes on an Apple M1, and 28 minutes on a hot 2019 Intel MacBook. See the benchmarks for details.
Measured, not promised
Results from our benchmark on clear, read English speech. Noisy audio, heavy accents and crosstalk give more errors. See the full benchmark.
Why your file never leaves your computer
Most online transcribers upload your recording to their servers. That costs them money for every minute, so free plans are capped at a few files or a few minutes a day, and your audio sits on someone else's computer.
Here the transcription happens on your own computer, inside this tab. Your recording stays on your device, so there is nothing to cap: transcribe as many files, as long as you like.
Questions
Is this the real OpenAI Whisper?
Yes: the open-source Whisper models OpenAI released, converted to run in a web browser. Nothing is sent to OpenAI or to us.
Does it work without an internet connection?
Once the model has downloaded, yes. Keep this page open, switch off your internet and transcribe another file at the same quality.
Can it translate into English?
No. It writes down what is said, in the language that is spoken.
Which quality should I choose?
Leave it on Auto. It picks the most accurate quality your computer runs well, and switches to a lighter one if your computer slows down. Best needs a fast computer.
Is it really free and unlimited?
Yes. Your device does the work, so a file costs us nothing to process. There is no daily cap, no length limit and no account.
Is my audio uploaded anywhere?
No. The file is read and transcribed inside your browser tab. The speech engine is downloaded once.
Why does it download something first?
That is the speech engine, which does the transcribing on your computer. It is a one-time download of 44–566 MB depending on the quality, and your browser keeps it after that. It has nothing to do with the size of your files.
Which browsers work?
Recent Chrome, Edge and Safari on a laptop or desktop. Chrome and Edge are usually fastest. Phones work for short files, but are slower.
Does it label different speakers?
Not yet. You get accurate text with timestamps; speaker labels are on our list.
Can I close the tab while it works?
Keep the tab open and in front. Browsers slow down background tabs, which can make a long file take up to four times longer.