Transcribe audio and video to text

No. The speech recognition runs inside your browser on this device, so your recordings, videos and transcripts are never sent to a server. The first time, your browser downloads the AI model from this site and keeps it; after that, nothing but the page is downloaded.

It knows 99 languages, and English, Spanish and Portuguese are among those it handles best. Clear speech comes out nearly word for word; expect some mistakes with names, strong accents, background noise or people talking over each other. More accurate makes noticeably fewer mistakes, above all in languages other than English. Translate to English turns speech in any of those languages into English text.

Up to 3 hours, or 30 minutes in browsers that can’t read sound piece by piece (such as Safari before version 26). On a recent computer, Standard usually takes a fraction of the recording’s length, and less with a graphics card; More accurate takes about three times as long, and phones are slower. Keep this tab open while it works; you can cancel at any time.

OpenAI’s Whisper (from “Robust Speech Recognition via Large-Scale Weak Supervision” by Alec Radford and colleagues), released under the MIT licence: whisper-base for Standard (a 137 MB download) and whisper-small for More accurate (250 MB), in the versions for browsers published by the ONNX Community on Hugging Face.

Yes. SRT and VTT files are plain text: open them in any text editor, or a subtitle editor, to fix a word or a time. Then add them to your video player, or upload them with your video to sites such as YouTube or Vimeo.