Transcribe audio or video
Turn speech into text or subtitles, on your own device. It is slow — genuinely.
- Runs on your device — the file is never sent anywhere
- No sign-up, no watermarks, no daily quotas
- Free, with no limit on how often you use it
How it works
Transcription services want your recording uploaded, and recordings are interviews, meetings, lectures and consultations — among the most sensitive files anyone handles. This runs the speech model on your own machine instead. The cost is time, and it is significant enough to state before you start rather than after.
How slow, realistically
The model is 41 MB and downloads once. After that, transcribing takes roughly as long as the recording itself on a modern laptop, and two or three times longer on an older machine or a phone. A five-minute clip is a coffee; an hour-long meeting is not practical. The length limit defaults to five minutes for that reason — raise it deliberately, not by accident.
What you get
Subtitles as .srt or .vtt with timings, ready to burn into a video with the subtitle tool here, or plain text for a transcript. Timings come from the model and are usually close but not frame-accurate.
Accuracy, honestly
This is the small Whisper model, chosen because the larger ones are hundreds of megabytes. It is good on clear English speech recorded close to a microphone. It is noticeably weaker on strong accents, two people talking over each other, background noise, technical terms and proper nouns. Punctuation is inconsistent. Read it through before sending it anywhere.
English only
The English-only model is used because it is smaller and more accurate on English than the multilingual version at the same size.
Transcribe audio or video — questions
How long will it take?
Roughly the length of the recording on a modern laptop, longer on older devices, plus a one-off 41 MB model download.
Is my recording uploaded?
No. The model comes to you and the audio is processed on your device.
How accurate is it?
Good on clear English speech, weaker on accents, crosstalk, noise and technical terms. Always read it through.
Can it do other languages?
No, this is the English-only model. Multilingual versions are much larger.