Transcribe video

Getting a good first pass

The transcript you get is decided almost entirely before you upload anything. If you still control the recording, the three things that matter most are microphone distance, room echo, and whether people talk over each other. If the recording already exists, upload the highest-quality copy you have rather than a compressed export — not because the resolution matters, but because re-encoding a re-encode degrades the audio.

Tell it the language if you know it. Automatic detection is good, but on a recording that opens with ten seconds of noise or music it occasionally guesses wrong and commits to that guess for the whole file.

Checking it without reading it twice

The fastest way to check a transcript is not to read it top to bottom. Play the audio and read along: errors announce themselves because the words on screen stop matching the words in your ears, and clicking a line jumps the audio to that exact second.

Errors cluster in predictable places, so check those first:

• ·Proper nouns — people, companies, products, places.
• ·Numbers, dates and units.
• ·The first few seconds, before the model has any context to work with.
• ·Any passage where two people overlap.
• ·Anything domain-specific: medical, legal, technical or internal jargon.

What automatic transcription is bad at

It is worth stating plainly, because the category tends not to. Automatic transcription is genuinely bad at overlapping speech, at proper nouns it has no way to know, at very heavy accents, and at distinguishing similar voices. It also produces confident, fluent, completely wrong text when the audio is poor, which is more dangerous than obvious gibberish because it reads fine.

That is why every line here is editable and why nothing on this site quotes an accuracy percentage. A number measured on clean studio audio would tell you nothing about your recording.

Turning the transcript into subtitles

Export SRT or WebVTT and you have a subtitle file that YouTube, Vimeo, LinkedIn and every video editor will accept. Scriptorium produces the subtitle file; it does not burn captions into the video itself, so the final step happens in your video tool or on the platform you upload to.