Transcribito

Transcribe an interview or meeting

For coursework, theses, journalism or minutes. You get the text with timestamps and can play back each sentence to check it.

1 · Audio

3 · Accuracy

4 · Transcript

  1. The model downloads once

    The first time, your browser downloads Whisper (80 to 150 MB in fast mode). After that it starts right away.

  2. It transcribes on your device

    The text appears bit by bit. A recent computer takes less time than the audio lasts.

  3. You fix it and download it

    Click a timestamp to hear that sentence, edit the text and download it as TXT or SRT subtitles.

Everything runs in your browser. Your audio is never uploaded: only the speech model is downloaded.

How to transcribe an interview

  1. Choose the recording: from your phone (M4A, MP3), a video call (MP4) or a voice recorder (WAV).
  2. For interviews with noise or several people, choose More accurate.
  3. Click Transcribe. You can read the first parts while the rest is processed.
  4. Check quotes by playing them from their timestamp, then download the text.

Good to know

  • It doesn't separate speakers. The text comes out continuously, with paragraphs where there are pauses. Mark who is speaking as you review it.
  • Always listen to quotes. Whisper is very accurate, but it can change a word or a name. Before quoting, play the sentence.
  • Long recordings. An hour takes anywhere from a few minutes (computer with WebGPU) to quite a bit longer than the recording itself (phone). Keep the tab open while it works.
  • Confidentiality. Since nothing is uploaded, you can use it with interviews with sources, internal meetings or conversations that must not leave your device.

Private by design

Most transcription services upload your audio to their servers. Not here: speech recognition runs in your own browser and the recording is never sent anywhere. That's why you can use it with confidential interviews, work meetings or personal voice notes.

Which model it uses

We use Whisper, OpenAI's speech recognition model, released as open source. It works in dozens of languages and handles different accents well.

  • Fast (Whisper base, 80 to 150 MB depending on your browser): good for voice notes, dictation and audio with one speaker and little noise.
  • More accurate (Whisper small, 250 to 500 MB): makes fewer mistakes with background noise, strong accents, several speakers or technical vocabulary. It takes longer, especially if your browser has no WebGPU.

The first time, the chosen model is downloaded and saved in your browser. Neither the audio nor the text is sent anywhere.

Other ways to use it

Frequently asked questions

Is my audio uploaded to a server?

No. Everything runs inside your browser. The only thing downloaded is the speech recognition model; your audio and the text never leave your device.

Is it free?

Yes. No sign-up, no watermark and no usage limits.

Does it tell who is speaking?

No. Whisper transcribes what is said but doesn't separate speakers. The text is split into paragraphs at pauses so it's easy to mark each person.

Can I use it with confidential recordings?

Yes. The recording is processed on your device and isn't sent to any server, not even ours.

How long does it take?

It depends on your device. On a recent computer with Chrome or Edge (WebGPU), fast mode transcribes much faster than the length of the audio. On a phone, or in browsers without WebGPU, it can take as long as the audio or longer.

Why does it need to download something?

Speech recognition needs a Whisper model (80 to 150 MB in fast mode and 250 to 500 MB in accurate mode, depending on your browser). It downloads only once and stays saved in your browser. That's what lets it transcribe without sending your audio to any server.

Which languages does it work with?

English and many others: Spanish, French, German, Italian, Portuguese, Dutch, Catalan… It's detected automatically by default; if you know it, you can choose it.

More apps I've made