Turn a lecture, an interview or a voice memo into text with speaker labels. It runs in your browser, so the recording never leaves your device.
- Free, no sign-up
- Nothing is uploaded
- One hour of audio in 15 to 30 minutes
English, German, French, Italian and 21 more European languages, and Chinese, Japanese, Korean and Cantonese. Works best on a computer with Chrome, Edge or Safari.
How it works
Choose a recording
A voice memo from your phone, an m4a, mp3 or wav file, or a video. Pick the language and how many people speak, or let the tool find out.
One download, the first time
Your browser downloads the speech models once: about 715 MB for European languages, 282 MB for Chinese, Japanese and Korean. After that it starts at once.
Your computer writes the text
One hour of audio takes 15 to 30 minutes, depending on the computer. Keep this tab open: the bar shows how long is left.
Name the speakers and download
Give the speakers names, tap a time to hear that part again, then download Word, text, Markdown or subtitles (SRT, VTT).
Privacy
The audio stays with you
The recording is turned into text inside your browser. It is never uploaded, not to this site and not to anyone else.
What is downloaded
Only the speech engine (from jsDelivr) and the speech models (from Hugging Face), the first time. Each file is checked against its SHA-256 fingerprint before use.
What is kept
The models stay in your browser for next time. The tool sets no cookies and only remembers your last choices. You can remove the models at the bottom of the tool.
Long lectures and interviews
- Use a computer. With 8 GB of memory or more, recordings of an hour or longer work. Phones often run out of memory with the European model.
- Plug in and keep the tab open. The tool keeps the screen on while it works, and the tab title shows how far it is.
- One speaker? For a lecture with one voice, choose “One, no labels” under Speakers. For an interview, “Detect” or the number of people.
- Good sound helps. Put the phone close to the speaker. Swiss spelling (ss for the sharp s) is an option for German.
- Private windows cannot keep the models, so they are downloaded again each time.
Checked on 20 October 2026. The speech models and their licences are named at the bottom of the tool. Only record people who agreed to it.