Transcript Cleaner

What the cleaner removes

A transcript copied from a video player is built for following along, not for reading: a timestamp every few seconds, a line break wherever the caption ended, [Music] in the middle of a sentence. The cleaner strips that scaffolding and gives you paragraphs. Every option can be switched off, and the result updates as you type.

Which formats it understands

• Our YouTube transcript — the copy with [mm:ss] Speaker A: before each segment. Segments of the same speaker are joined, and the speaker label is written once per paragraph.
• YouTube's own transcript panel — the time on one line, the words on the next. Time lines are dropped, including the spoken form some copies carry ("1 minute, 5 seconds").
• SRT, WebVTT and SBV subtitle files — cue numbers, timing lines, VTT headers, notes and inline tags go; a VTT voice tag becomes a speaker label; the repeated line at the top of every auto-caption cue is removed.
• Meeting notes with a name and a time on their own line ("Speaker 1 0:03") become "Speaker 1:".

Only timestamps that look like timestamps are removed: at the start of a line, on a line of their own, or in brackets. "We meet at 10:30" in the middle of a sentence stays as it is.

How paragraphs are rebuilt

When the text came from a player, every line break is treated as the end of a caption rather than a paragraph, so the lines are joined first. A word split across two lines with a hyphen is put back together. Then a new paragraph starts every 3, 4, 5, 6 or 8 sentences (4 by default), and always when the speaker changes. Text you typed yourself keeps its blank-line paragraphs.

Auto-generated captions usually have no periods at all, so there are no sentences to count. When a text averages more than 60 words per sentence, the cleaner cuts paragraphs by length instead — about 20 words for every sentence you asked for — and tells you so.

"Fix spacing and capitals" removes double spaces and spaces before commas and periods, adds a missing space after a comma, and capitalises the first letter of each sentence. Abbreviations such as "Dr." and initials such as "U.S.A." do not end a sentence, and words like "iPhone" keep their first letter.

Filler words, carefully

The default lists hold hesitation sounds only — "um", "uh", "erm" in English, "э", "эм" in Russian, "äh", "ähm" in German — for ten languages: English, Russian, Ukrainian, Spanish, Portuguese, French, German, Italian, Polish and Dutch. The page language picks the list. Words such as "like", "you know" or "o sea" are left out on purpose: they are also ordinary words, and removing them blindly changes meaning. Add them to the list yourself if you want them gone; phrases of several words work too, and your list is remembered in this browser.

Honest limits

The cleaner tidies text, it does not understand it. It cannot add punctuation that the source never had, cannot fix misheard words, and a line that starts with a capitalised word and a colon ("Note: …") can be mistaken for a speaker label — that option is off by default. Read the result before you publish it. Everything runs in your browser; the text is not uploaded or stored.