Transcribe an interview

1. 1Choose your an interview recording.
2. 2Pick the language, or leave it on Detect automatically.
3. 3Press "Transcribe". Fix any word, then copy the text or save it as TXT, SRT or VTT.

Click the time next to any line to hear exactly that part, then fix the words in the page. Do this for every quote you publish: names, numbers and words that sound alike are where a speech model slips.

Your interviewee's words are never uploaded to a transcription service. That keeps an off-the-record recording on your own device.

• The text, split where the speaker pauses. Fix any word right in the page.
• Times for every line. Click a time to hear that part again.
• Copy it, or save it as a text file (TXT) or as subtitles (SRT or VTT) for a video player or YouTube.
• Translate to English: speech in another language comes out as English text.

The first time, your browser downloads the speech model (145 MB). It is kept for next time, so later recordings start at once.

In our test, a 96-second lecture took about 5 seconds on a Mac laptop in a Chrome-based browser, using the graphics chip. Without one, the same laptop took about 32 seconds. Phones are slower. You see the text appear while it works.

Clear speech in English gives the fewest mistakes. Other languages work too, with more mistakes; names and rare words are the usual errors, so read it through before you share it.