Where the words come from
This tool does not listen to the soundtrack. It asks YouTube for the caption tracks already attached to the video, picks one, and downloads it as timed events. If a video has no caption track, there is nothing for us to convert.
YouTube's help pages say automatic captions are made with speech recognition and that their quality can vary with pronunciation, accents, dialects or background noise. Uploaded captions are written by the creator. In every track list we pulled on 17 September 2026, automatic tracks were marked asr and named "auto-generated"; uploaded tracks were not.
Uploaded captions vs automatic captions
On a video with only an uploaded English track, the caption file had 6 events and 222 characters. No line repeated and no cue overlapped the next, so the caption file held six short, separate lines.
On a video with only an automatic English track, the file had 43 events. Eleven were exactly [Music]. Three lines were identical to the line right after them. The first cue started at 1,280 ms and lasted 7,040 ms; the second started at 4,960 ms, so the two overlapped by 3,360 ms. That is a rolling caption: the player keeps old words on screen while new ones are added, and the file stores the same phrase again.
A second automatic English track had 42 events, 10 lines of [Music], and one 18-character question stored 7 times.
How the lines become paragraphs
We keep every caption event YouTube returns, in order. With timestamps on, you see one line per event with its start time. With timestamps off, and in Copy and TXT, the events are joined into paragraphs: line breaks inside an event become spaces, empty events are dropped, and an event that repeats the one right before it is kept once.
Once a paragraph passes about 250 characters, a new one starts after the next sentence ends or after a pause of a second and a half or more. Captions with no punctuation are broken at about 600 characters. We do not repair punctuation, label speakers or merge overlapping windows. SRT and VTT keep every event with its own start and end time.
TXT, SRT or VTT
TXT is the paragraph text, without timestamps: right for notes, documents and quotes. SRT and VTT keep each event's start and end time for subtitle editors, video timelines and web players; overlapping automatic cues stay overlapped. The download page goes through the files in detail.
Which language you get
There is no language picker. When the video has an English track we take it; otherwise we take the first remaining track. One video with five uploaded tracks and no English came back in Indonesian (id-ID, 2,509 characters).
We do not request YouTube's automatic translation. When we tried its translation parameter on 17 September 2026, YouTube answered HTTP 429, so we have no translated file to show you.
Why some videos return no text
- No caption track. Three playable videos we tested on 17 September 2026 had zero caption tracks. The message says so. If it is your own video, you can upload the file instead.
- YouTube asks our server to sign in. Sometimes YouTube treats our request as automated and will not share captions, even for public videos that have them. On the evening of 19 September 2026, a batch of 10 captioned videos all came through; a second batch of 10, about ten minutes later, was refused every time. Direct requests are intermittent. We ask YouTube directly first; if it declines, we retry once through a relay server we operate. If it still declines, we show you how to copy the transcript from YouTube yourself.
- Live replays. One live recording came back unplayable ("This live stream recording is not available.") with no caption track.
- Music. Music-heavy videos still had automatic captions, just noisy ones: [Music] in 11 of 43 lines on one track and 10 of 42 on another. A Korean music upload returned 67 events.
- Region. Asking for the same automatic-only video as the US, Japan, Brazil and Germany gave the same single English track each time. We could not reproduce a region-based gap.
- Turning CC off in your own player changes nothing here, because we do not play the video or load an embedded player.
Where the text goes wrong
- Music: lyrics around [Music] tags, with repeated words.
- Accents and messy speech: filler and cut-off phrases, such as a line ending in "uh yeah" and an event that was only "caught in a".
- People talking over each other: YouTube says overlapping speech can stop automatic captions from appearing. We have not isolated a clip that proves it.
- We have not scored any track against a human transcript, so we do not publish an accuracy percentage.