Transcript cleaner
Paste a transcript, or open a .txt, .srt or .vtt. Take out filler words, timestamps and cue numbers, and join each speaker's turns. It runs in your browser — the text never leaves it, and there's no account.
What it does, and the rules it follows
Every change is a rule you can check against the result box. Where a rule has to guess, it guesses towards leaving a word in.
- Filler words
- Sounds with no other meaning — um, eh, hum, euh, äh and the like — go wherever they stand. Words that are fillers only sometimes go only where punctuation sets them off, and only where they are seldom anything else: some wherever commas set them off (“we, like, went”), some only opening a sentence (“Well, I think”), some only closing one (“c’est cool, quoi.”). A word that is a sentence on its own (“Right.”) stays: it is an answer. The word that then opens the sentence is capitalised; a filler written in capitals inside lower-case text is taken for an acronym and kept. The lists are under Questions below.
- Speaker names
- A “Name:” at the start of a line is a speaker when it opens at least two lines, when it reads “Speaker 1”, when a time stands between it and its colon (“Maya (00:01:23):”), or when the text is written as “Name: text” lines and some name recurs. Otherwise a word that opens one line only — “Warning:”, “Note:” — stays as written, because dropping it would be worse than keeping a name. A name starts with a letter, so a score such as “3:2” is never one. A WebVTT voice tag always names its speaker.
- Timestamps and cue numbers
- A time at the start of a line — [00:01:23], (1:30), 0:02:05 — or after the name — Maya (00:01:23): — is the paragraph's time. A line that is only a name and a time, as meeting apps export, names the paragraph under it when it reads “Speaker 1” or the name heads more than one such line; otherwise it stays as text, since “Lunch at 12:30” looks the same. A time inside a sentence (“we met at 12:30”) is left alone. Cue numbers go with the timestamps.
- Subtitle files
- SubRip and WebVTT are read as subtitles, each cue a paragraph with its line breaks joined; a line break is never taken for the end of a sentence. With the timestamps kept, the cleaned cues can be downloaded as .srt or .vtt with their timing unchanged and the same words as the result box; merging turns applies to the text only, since a cue keeps its own time on screen.
- Merged turns
- Consecutive paragraphs by the same named speaker become one, carrying the first one's time. A paragraph with no name is never joined to anything: whether it continues the speaker before it is a guess, and it reads the same either way.
- Byte-order marks and Windows line endings
- Stripped and normalised, rather than turning the first cue into an unreadable block.
Questions
Is my transcript uploaded?
No. The text is read and rewritten on this page, in your browser, and none of it is sent anywhere. The link at the bottom leads to txscribe, which transcribes recordings in the cloud; this page is not that.
Which filler words does it remove?
The ones in its lists for English, Spanish, Portuguese, French and German — pick the language above the text; each language's words are set apart by a semicolon. Sounds with no other meaning go wherever they stand: um, uh, uhm, er, erm, ah, eh, hm; eh, ehm, em; hum, hã, ahn, ãh, eh; euh, hum, hm, bah; äh, ähm, öhm, ehm, hm, mh. These go wherever punctuation sets them off on both sides: like, actually, basically, literally, you know, i mean, sort of, kind of; o sea, digamos, en plan, pues; né, quer dizer; en fait, du coup, tu vois, hein; sozusagen, quasi, ich meine, irgendwie, halt. These go only opening a sentence, with a comma after them: so, well, right, okay, ok, anyway; este, bueno, mira, oye, tipo, a ver; então, assim, bom, enfim, olha, tipo, é; bon, alors, donc, enfin, voilà, bref, genre; also, naja, na ja, nun. These go only closing a sentence, after a comma: sabe, entendeu; quoi, voyez; gell, weißt du, weisst du. None of them goes when it is a sentence on its own: that is an answer.
Why was a word kept that I’d call a filler?
Because nothing set it off, or it stood where it is often a real word. “We like went home” keeps its “like”: without a comma on either side the page cannot tell it from “we like going home”, and the rule is to leave a word in rather than take meaning out. Add the commas and it goes.
Can I keep the timestamps?
Yes: untick “Strip timestamps and cue numbers”. Each paragraph then opens with its time, and when the input was a .srt or .vtt the cleaned cues can be downloaded as .srt or .vtt with their timing unchanged.
What does it read?
Plain text with “Name: text” lines, times at the start of a line or after the name, meeting-app headers like “Speaker 1 0:03”, SubRip and WebVTT files, WebVTT voice tags, Windows line endings and a byte-order mark. Audio and video files are refused: this page cleans text, it does not transcribe.
Is it free?
Yes, with no account and nothing to install.
Need the transcript first?
txscribe transcribes audio and video with a timestamp on every word, and exports plain text, SubRip and WebVTT. Unlike this page, that part runs in the cloud.
Start free