How does sayit detect pauses and filler words like 'um' and 'like'?

sayit listens for two separate things that both affect how fluent a take sounds: filler words like "um" and "like" scattered through your speech, and silences that run long enough to disrupt the flow. Both feed into your fluency score, but neither is judged in a vacuum — a natural pause at the end of a sentence is treated very differently from the same length of silence in the middle of a phrase.

30-second version: sayit matches nine common filler markers (um, uh, er, like, you know, literally, basically, actually, so) against your transcript and counts them. Separately, it measures how much of your recording was silence, gives you a free allowance for the pausing a natural reader does at sentence and clause boundaries, and only charges the excess. Both filler density and excess silence shave points off fluency — gently, and only past a point.

What exactly counts as a filler word?

sayit matches a fixed set of markers against your transcript, case-insensitively and with punctuation stripped: um, uh, er, like, you know, literally, basically, actually, and so. "You know" is matched as a two-word phrase across consecutive words rather than needing "you" and "know" to each trigger separately. These are the words that carry no real content but signal hesitation or a verbal tic — and detecting them requires a real transcript, which is why filler counting only happens where speech-to-text runs (freeform speech and, when enabled, reference mode).

Does every filler word tank my score?

No — filler density shaves a bounded amount off your rate score, deliberately capped so a handful of "um"s in a longer answer doesn't wreck an otherwise fluent take. The point isn't to punish every hesitation marker into oblivion; it's to reflect that a take dense with filler words reads as less fluent than one without, in proportion, not as an instant failure.

How does sayit tell a "good" pause from a "bad" one?

By checking what came right before the silence. A gap that follows a full stop, comma, or other clause-ending punctuation gets a free allowance — up to about a second of silence there costs nothing, because that's exactly where a natural reader is supposed to breathe. Silence that shows up somewhere else, or that runs past that free allowance even at a real boundary, starts counting against you. On top of that, sayit prices the share of your whole read that was long silence (pauses over roughly 0.3 seconds): up to about a tenth of a read can be pausing, for free, before the excess starts costing fluency points.

Why does this matter — can't you just count "long pauses"?

Because a punctuation-blind pause count contradicts the pacing feedback shown right next to it. The gap before "…and" after a long clause is exactly the kind of place a competent reader is supposed to pause — the tempo and pacing analysis elsewhere on the same results page already treats a sentence-end gap as expected, not a fault. Charging that same gap a flat penalty in the fluency score would mean two numbers on one screen disagreeing about the same breath, which is confusing rather than informative. Giving boundary pauses a free allowance keeps the two scores reading the same gap the same way.

What still counts as a long pause, even at a boundary?

sayit still records every gap over about 0.6 seconds as a "long pause" in the descriptive breakdown, whether or not it happened at a natural boundary — a boundary breath is still a long pause, it's simply not treated as a fault against fluency. That distinction matters if you're trying to diagnose why your fluency score moved: the long-pause count tells you where the silence was, the fluency penalty tells you how much of it was actually excessive.

Pauses and fillers at a glance

Try it

Read a few sentences with natural punctuation in sayit and check your fluency breakdown — you'll see filler count and long pauses reported separately, with the pacing card right alongside showing which pauses were expected and which weren't.