Transcript cleaner no timestamps, no labels, in paragraphs.

How do I remove speaker labels and [Music] tags?

Meeting, interview and podcast transcripts usually open each turn with a label: "Speaker 1:", "SPEAKER_00:", "Interviewer:" or the person's name. You can remove every label, remove only the generic ones (and keep real names) or keep them all. Even when a label goes, a change of speaker still starts a new paragraph, so the conversation does not blur together.

To avoid deleting real text, a name only counts as a label when it is generic, when it opens more than one turn or when most lines carry one. A single "Note: bring a dish" inside ordinary text stays where it is.

Sound tags go too: anything in square brackets, like [Music] and [Applause], parentheses that describe a sound, like (laughs), (coughs) and (inaudible), and song lyrics between ♪ marks. Parentheses with real content, like (2019) or (see chapter 3), stay.

How do I join broken subtitle lines into paragraphs?

Subtitles are made for the screen: each cue holds two short lines and the sentence carries on in the next cue. To read, edit or publish it, that has to become running text. The paragraph option joins every line and starts a new paragraph every few sentences (four by default, you choose). A blank line in the original and a change of speaker always start a paragraph.

Auto-generated captions often come with no punctuation at all. Then there are no sentences to count, so paragraphs are cut by length, about twenty words for each sentence you asked for. You can also ask for one sentence per line, handy for reviewing or building a script, or keep the lines as they are and only strip the clutter.

The last option handles the finish: it removes double spaces and spaces before commas and periods, drops the comma left behind when a hesitation goes, and capitalizes the start of each sentence (and a lone "i" in English). Abbreviations such as Dr. and p. are not mistaken for the end of a sentence.

Can I remove "um" and "uh" from a transcript?

Yes, with the hesitation option, which starts off. It removes only sounds that carry no meaning in any sentence: in English, um, uh, erm and hmm; in Portuguese, ahn, hum and é... (with the dots, since é alone is a verb); in Spanish, eh, mmm and este... with the dots. The language is detected from the text itself.

Words that are sometimes a filler and sometimes not, like "like", "so", "you know" and "actually", stay. Deleting them all would change sentences such as "I like this light". To see how often they show up, use the filler word counter.