Tools

Filler word counter

Paste a transcript and see where the hesitations are. Every match is marked in place, counted by term, and totalled, so you can tell a habit from a one-off before you decide to cut anything.

The counter runs entirely in your browser - nothing is uploaded, nothing is stored, and nothing you type leaves the page.

The marked-up transcript appears here once there is one to mark.

What it counts

The hesitation sounds - um, uh, er and ah - and the phrases that do the same job: you know, I mean, sort of, kind of, basically, actually, literally, a question that ends in right, and so at the start of a sentence. Two of those are marked as heuristics rather than counted plainly. Like is usually a verb or a comparison, and a sentence can legitimately open with so, so both are shown faintly and labelled instead of being folded into the total as though they were certain.

Matching is whole words only, so unlike and likeness are left alone, and overlapping phrases are counted once: sort of is one match, not a match plus whatever sits inside it. The seconds figure assumes 0.4 seconds for each filler, which is an assumption and is labelled as one everywhere it appears.

Removing them, rather than counting them

Counting is the easy half. Rooster’s planned filler-word removal is review-before-apply: the editor lists each instance with the words around it and its timestamp, and you accept the ones you want gone rather than handing the whole transcript to a model. It runs on your Mac, so it is not metered and there is no per-action credit, and cuts snap to measured silence rather than to a transcript timestamp. It is in development and not released yet - the counter above is text matching in a browser tab and is not the same thing. What is planned, and what already works, is on the features page.