Remove filler words from a transcript

Verbatim and clean-read transcripts

A verbatim transcript keeps everything that was said: the ums, the false starts, the repeated words. That is what you want for research, for a legal record or for anything where how somebody spoke matters. A clean-read transcript removes those so the text reads like writing. Most articles, show notes and meeting summaries want the second. This tool turns the first into the second without retyping.

What it removes

• Always: um, umm, uh, uhh, uhm, er, erm, hmm, mm, mhm. These are not words in any sentence, so they are always safe to drop.
• Optionally, between commas: like, you know, I mean, sort of, kind of, basically, actually, literally. Only when a comma comes right before and right after, which is the way a transcript marks a word being used as a filler.
• Optionally, repeats: a word said twice or more in a row (“the the”, “I I think”) becomes one. A repeat on purpose, such as “very very good”, is also collapsed, so turn this off if that matters to you.

What it repairs around the gap

Um, so today I want to talk about, uh, how we plan the quarter.
It was, like, a lot of work, and I I think the the plan is solid.

So today I want to talk about, how we plan the quarter.
It was a lot of work, and I think the plan is solid.

When a filler opens a sentence, the next word gets its capital letter. When a filler closes a sentence (“went there, uh.”) the full stop moves back onto the word before. Spaces and stray spaces before punctuation are tidied. One thing it does not do is judge a comma: the comma left in “talk about, how we plan” is the speaker's, and only you can say whether the sentence needs it.

What it cannot do

It cannot tell a filler from a word. That is why the risky words are limited to comma-set-off cases, and why nothing here removes “well”, “so” or “right”, which are sometimes fillers and sometimes the point. It does not rewrite grammar, and it does not remove false starts (“I went to the, the, I went home” keeps the first clause apart from the collapsed repeat). Read the result once before you use it.

Check what changed

The note under the result says how many fillers were removed, how many repeated words were collapsed and how the word count moved, so you can tell at a glance whether the tool did a little or a lot. A transcript of unscripted speech often loses several percent of its words, and an interview full of hesitations more. If the number looks too large, turn off the repeats or the comma-set-off words and compare.

Where this fits

Do the clean-up last. First remove the timestamps if the transcript has them, then run the filler remover, then check the length with the word counter. If you have the recording rather than the text, a Scriptorium transcript is a verbatim-leaning draft that you can export as plain text and then clean here.