Private audio transcriptionTranscribe a recording without uploading it.
Drop in an audio file, or record in the page, and a speech model running in your own browser turns it into text. The recording is never sent anywhere — not to us, and not to a transcription service.
How it works
The model comes to you
The first time, a speech model of about 80 MB loads into your browser from us and is kept there. Nothing is installed. From then on, transcription runs on your own device.
Your audio stays where it is
A file is read in the tab. A recording made in the page is transcribed as you speak, a sentence or so behind, and kept in your browser beside the text so you can play it back. Neither is uploaded.
Then use the text
Timestamped lines, a search box over them, copy, and download as .txt. Ask for a summary and it goes to the chat model you have chosen — in your browser, or encrypted to our hardware if that is what you picked.
What it does well, and what it doesn't
- Best on clear speech, one speaker, in English. It uses Whisper base, a small model. Accents, crosstalk and background noise make it worse, and quality falls off outside English.
- No speaker labels. You get timestamped lines of what was said, not who said them.
- Unclear passages are marked, not guessed. Where the model loops or returns nothing, the line reads [unclear]. It is a rule about what the model did, not a confidence score, so it will not catch a fluent wrong word.
- You can name the language. English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Russian, Japanese or Chinese — or leave it to be detected.
- Speed depends on your device. On a recent laptop a minute of audio takes roughly ten to fifteen seconds. An older machine or a phone is slower, and if live transcription falls behind it says so and keeps every word.
- No limits on length or minutes. Files up to 200 MB, because the decoded audio is held in memory while it is read.
Want to check that the audio stays on your device, rather than take our word for it? How to transcribe a recording without uploading it shows two ways.
Questions worth asking
No. The speech model is loaded into your browser and runs on your own device, so the recording is read in the tab and turned into text there. It is not sent to us or to a transcription service, whichever mode the chat is in. The model itself, about 80 MB, is downloaded from us once and kept in your browser.
It's free, needs no account, and has no minute, monthly or file-length limits, because the work is done by your own device and costs us nothing. Files up to 200 MB can be read; the limit is there because the decoded audio is held in memory.
It uses Whisper base, a small speech model. It's strongest on clear speech, one speaker at a time, in English. Accents, people talking over each other and background noise make it worse, and quality falls off outside English. Where it loops or returns nothing, the line is marked [unclear] rather than guessed. Check anything you rely on against the recording.
No. It doesn't label speakers. The transcript is timestamped lines of what was said.
You can name the language as English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Russian, Japanese or Chinese, or leave it to be detected. Naming it helps when detection guesses wrong, because a wrong guess produces fluent text in the wrong language rather than obviously broken text.
In your browser, on your device, unless you switch saving off. A recording made in the page is kept beside it so it can be played back or downloaded. You can copy the text, download it as a .txt file, export everything to a file, or delete it. Nothing is kept on our side.
The transcript is handed to whichever chat model you have chosen. If that model runs in your browser, the transcript stays on your device. If you've chosen our hardware, the transcript is encrypted in your browser and sent there to be summarised, and it isn't written down. The audio never goes anywhere in either case.