How to transcribe a video

A transcript without sending your recording to a transcription service — which matters when the recording is a meeting.

Most transcription services work by uploading your recording. For a lecture that may be fine; for a client call, an interview or an internal meeting it is often not acceptable. This runs the recognition model in your browser instead.

Step-by-step

1. Add your video or audio file.
2. Wait for the model to download the first time. It is cached afterwards.
3. Transcribe. Expect this to take a while — see below.
4. Review and correct, then export.

What to expect on speed and accuracy

Running locally is slower than a data centre. A one-hour recording takes a substantial fraction of an hour on a typical laptop, and the tab needs to stay open. Start it when you do not need the machine.

Accuracy is good on clear speech and degrades with background noise, crosstalk and strong accents — as all speech recognition does. Treat the output as a draft to correct, not a finished transcript. Names, technical terms and numbers are where errors cluster, and they are exactly the parts people quote.

Getting better results

• Better audio beats better software. A close microphone helps more than any setting.
• Avoid crosstalk. Overlapping speakers are the hardest case by a wide margin.
• Trim first. If you only need ten minutes of a two-hour recording, cut it before transcribing.

Frequently asked questions

Is my recording uploaded?

No. The recognition model is downloaded to your browser and runs there. The recording never leaves your device.

How long does it take?

Considerably longer than a cloud service, since it runs on your own hardware. Budget a significant fraction of the recording's length and leave the tab open.

How accurate is it?

Good on clear, single-speaker audio; noticeably worse with background noise or overlapping speech. Always review the result, particularly names, numbers and technical terms.