Video Transcription: A Practical Guide
What video transcription is, why it is worth doing, how automatic and human transcription compare, and what to do with the transcript once you have it.
The Scriptorium team ·
Video transcription is the process of turning the speech in a video into written text. The result is a transcript: a readable record of what was said, usually with timestamps, and with speaker labels when more than one person talks.
Why transcribe a video at all
Video is a bad format for finding things. You cannot skim it, you cannot search it, and you cannot quote from it without watching it again. A transcript fixes all three, which is why the reasons to make one come from very different places:
- Accessibility. Deaf and hard-of-hearing viewers need captions, and captions come from a transcript. In many contexts this is a legal requirement.
- Search. Search engines cannot watch a video. A transcript on the page is text they can index.
- Repurposing. A transcript is the raw material for show notes, articles, summaries and social posts.
- Reference. Finding the two minutes that mattered in a ninety-minute recording takes seconds with text and ages without it.
- Comprehension. Many people read faster than they listen, and many prefer to.
Automatic versus human transcription
| Automatic | Human service | |
|---|---|---|
| Turnaround | Minutes | Hours to days |
| Cost | Free to a few currency units per hour | Priced per audio minute |
| Clear single speaker | Very good | Very good |
| Crosstalk and noise | Weak | Good |
| Specialist vocabulary | Weak without help | Good with a brief |
| Effort from you | Checking and correcting | Little |
The practical answer for most people is automatic first, then correct it. Human services earn their price on legal proceedings, medical records, publishable research, and recordings where several people talk over each other.
How automatic transcription works, briefly
A speech recognition model is trained on a very large amount of recorded speech paired with its text. Given new audio it predicts the most likely sequence of words. Modern models use surrounding context, which is why they punctuate sensibly and usually pick the right one out of a set of homophones.
It also explains the characteristic failure. Because the model predicts likely text, unclear audio does not produce obvious errors — it produces confident, fluent, plausible sentences that are wrong.
How accurate is it?
Nobody should quote you a single accuracy figure, including us. Automatic transcription of one person speaking clearly into a close microphone is very good — good enough that correcting it is much faster than typing it. The same system on a recording with two people talking over each other, a noisy room, a strong accent or a lot of specialist vocabulary is meaningfully worse.
The failure mode worth knowing about is that it does not degrade into obvious nonsense. It produces fluent, confident, plausible sentences that are wrong, which is harder to catch than gibberish would be. That is why you check a transcript against the audio rather than reading it on its own.
What to do with the transcript
- Captions. Export SRT or WebVTT and upload it alongside the video.
- An article. A transcript is a draft with terrible structure. Reorganising one into a readable piece is much faster than writing from nothing.
- Show notes and chapters. Timestamps give you the chapter marks directly.
- A searchable archive. The real payoff arrives once you have transcribed enough that you can search across all of it.
- Quotes. With timestamps, verifying a quote is one click rather than a hunt.
Getting started
If you have a file to hand, video to text takes it directly — free for a first file up to 15 minutes, with no account. If the video is on YouTube, start at the YouTube transcript page instead, which covers the free route for videos that are not yours.