Audio to Text Transcription

It uses Whisper (MIT licence), OpenAI’s open speech recognition model. Under “Where it runs” you choose: our server, where Whisper large v3 writes it in seconds with nothing to download (the sound is sent in pieces and not stored), or this device, where nothing is uploaded and a smaller Whisper downloads once (40 to 240 MB, depending on the accuracy you choose). More about the server.

On this device: Balanced suits clear speech in common languages. Best makes fewer mistakes with accents, noise and less common languages, but downloads more and takes longer. Fast is for a quick rough draft.

On our server, a few seconds per minute of recording, with 20 minutes a day free. On this device, usually less than the recording lasts on a computer, and longer on phones. Text appears as it’s written, and up to 60 minutes of a recording are transcribed (20 on phones).

No. It writes down what is said, in paragraphs where the speech pauses, but it doesn’t tell speakers apart. Names and specialist terms may need correcting.

Yes: save the result as SRT or WebVTT here, or use the Subtitle Generator, which shows the subtitles as it writes them. Add Subtitles to Video burns them into the picture.

It is free, with no watermarks and nothing to sign up for. On this device there are no daily limits. Our server has a free daily allowance, and a free account makes it six times bigger.