How to turn a transcript into an SRT subtitle file
A transcript and a subtitle file contain the same sentences, but a subtitle file also answers two questions a transcript never asks: where does each caption end, and how long does it stay on screen. Getting those two right is most of what separates subtitles that disappear into the video from subtitles that are work to read.
First, strip the scaffolding
Dictation and meeting tools wrap their output in furniture that has no place in a caption: their own timestamps at the head of each paragraph, a speaker label on its own line, blank-line padding, sometimes a running line number. All of it comes out before anything else happens, because a caption reading "Speaker 1 So I went down to the shop" is worse than no caption. Paste the transcript as it came and the tool removes the timestamps, the speaker headings and the stray numbering on the way in.
Where to break each caption
The single biggest quality difference is where a caption ends. A full stop is the best place — the viewer finishes a thought and the screen clears. A comma, a semicolon or a dash is the next best. Breaking in the middle of a clause is what makes amateur subtitles tiring: the eye reaches the end of the caption with the sentence unresolved and has to hold it until the next one appears. So the order of preference is: end at a sentence if the sentence fits, break at a clause boundary if it does not, and only fall back to a plain word boundary when there is nothing better within reach. That is the order the tool works in.
How long each caption stays up
Duration should follow the amount of text, not a fixed value. The usual unit is characters per second: roughly 17 is a comfortable adult reading speed, and captions aimed at younger viewers are usually slower. Two limits sit on either side of it. A very short caption still needs about a second on screen, or it reads as a flash rather than a line. And a caption held much beyond six seconds starts to look like a stuck subtitle even when the text justifies it — at that point the right answer is usually to split it. A small gap between consecutive captions matters more than it sounds. Without one, two captions can read as a single block of text that mysteriously changed; a couple of frames of clear screen is enough to mark the boundary.
Then line it up with the audio
Timing a transcript from nothing gives you captions with the right rhythm relative to each other, but no anchor to the actual speech — the first line starts when you say it starts, and everything follows from there. In practice that is close enough for a video where the talking begins immediately, and needs one adjustment where it does not. Set a start offset for the leader or the titles, then play it back and shift the whole file if it needs it. If the speech has long silences in the middle, you will get better results splitting the transcript into a few chunks and timing each one against its own starting point.
Check it before you ship it
Two numbers are worth a glance. The longest line: if it exceeds the width you set, something did not wrap. And the count of captions running faster than your reading speed: a handful is normal on dense speech, but if most of the file is flagged, the reading speed you set is not the one the transcript needs. Then watch it back once, at speed, without pausing. Anything that makes you stop is a caption to fix.
Frequently asked questions
Can a tool know when each word was spoken?
Not from the text alone — a transcript carries no timing information. What can be done is to distribute the words at a realistic reading speed and let you set where the first caption starts, then shift the whole file to line it up with the audio.
What if my transcript already has timestamps in it?
They are usually coarse — one per paragraph or per speaker turn — and not accurate enough to caption from directly. Those timestamps are stripped on the way in, and the captions are timed from the text.
How many characters should a caption line hold?
42 is the usual maximum for Latin-script subtitles, over at most two lines. Longer than that and the eye travels too far; the line also risks being wrapped by the player itself, in a place nobody chose.
Should I keep the speaker names?
In a caption file, normally not — the viewer can see who is speaking. They are worth keeping when speakers are off-screen or hard to tell apart, in which case the convention is a short name in capitals at the start of the caption.
More guides
Other things we made
Also in English, made by the same people.
Also in English, made by the same people.
Also in English, made by the same people.
Also in English, made by the same people.