VTT to text converter

Why an auto-caption transcript reads badly

Automatic captions are written for a rolling window rather than for reading. Each cue restates the previous one and adds a few words, so the text scrolls smoothly on screen — and pasting a YouTube VTT straight into a document gives you the same phrases over and over.

Removing repeated lines is on by default. It drops a cue that simply repeats the one before it, and where a cue restates the previous line and extends it, it keeps only the longer version. Turn it off if a line is genuinely said twice in your file.

It will not try to stitch together cues that only partly overlap, such as "we are going" followed by "going to the shops". Working out where to join those means guessing, and a wrong guess deletes words from your transcript silently. Anything it cannot resolve is left in place for you to see.

One line per cue, or flowing paragraphs

Subtitle cues break where the screen runs out of room, not where a sentence ends, so a faithful line-per-cue transcript is full of sentences chopped in half.

Merging into paragraphs joins the cues back up and starts a new paragraph after a full stop, question mark or exclamation mark. That is the version you want for reading, quoting, or pasting into something that summarises it. Keep the line-per-cue version if you plan to line the text back up against the video.

What gets removed

Timestamps, cue identifiers, cue settings and NOTE blocks all go. So does WebVTT markup: voice spans, class spans, karaoke timing tags and the character references the format uses for angle brackets and ampersands, which otherwise show up as & in the middle of your text.