Where the SRT timestamps come from
The times in the file are not guessed from the soundtrack. Each caption event arrives from the caption track with a start time and a duration in milliseconds. Your browser writes the start as HH:MM:SS,mmm and the end as start plus duration, or start plus 3,000 ms when the duration is missing.
On one track we pulled through this site, the first event started at 80 ms and lasted 3,760 ms. The SRT cue this page writes for it is 00:00:00,080 --> 00:00:03,840, because 80 + 3,760 = 3,840 ms.
On the same file, 8,614 of 8,721 cues overlapped the next cue: one cue ended after the next one had started. That overlap stays in the download. We do not clean it up before saving.
How big a long transcript file gets
We ran this site's caption request on a long captioned upload and built both files with the same rules the download buttons use:
- 8,721 caption events
- TXT: 295,849 bytes, 17,441 lines
- SRT: 599,978 bytes, 34,884 lines
- First cue starts at 80 ms; the last cue starts at 19,325,280 ms, 5.368 hours into the video
SRT was larger than TXT because every cue repeats a number line and a time line. Two more captioned uploads, built the same way in the same run:
- 4,953 events: TXT 160,776 bytes / 9,905 lines; SRT 333,025 bytes / 19,812 lines
- 2,945 events: TXT 112,648 bytes / 5,889 lines; SRT 214,617 bytes / 11,780 lines
These are three observations, not an average, so we do not quote a typical file size.
Copy or download
Copy puts the TXT text on your clipboard and creates no file. Downloads are built in your browser: .txt as text/plain, .srt as application/x-subrip and .vtt as text/vtt, each named after the video ID.
SRT times use a comma before the milliseconds, as in 00:00:00,080. VTT uses a period and starts with a WEBVTT line. SRT cue numbers start at 1 and count up in file order.
When there is nothing to download
The download buttons sit in the result, which only opens when a transcript arrives. If the request fails, there is no file to save, and the status line says why.
If YouTube refuses the request, the page offers the copy-from-YouTube route: open the video on YouTube, copy its transcript and paste it here. A pasted transcript downloads the same way.