Filler Word Counter
Speak for 30 seconds and get a straight count of your ums, uhs and likes — plus what that works out to per minute. Those three words are the only ones this counts. To read them, your clip is sent to a transcription service and turned into text; we keep neither the audio nor the transcript.
Talk for 30 seconds
Talk the way you would to a friend — what you did this morning, how your week is going. Do not read anything out: fillers show up when you are working out what to say next.
Your 30 seconds are sent to Deepgram, our transcription provider, to be turned into text. We don't keep the audio or the transcript.
What happens to your recording
When you tap record, your browser captures about 30 seconds of audio. It first checks that there is speech in the take at all: a silent one fails right there, and no clip is sent anywhere. Otherwise the clip goes over an encrypted connection to this site and on to Deepgram, the speech-to-text provider we use, which returns the words you said. The counting happens after that, on the text.
We do not save the audio and we do not save the transcript. Neither is written to a database, put in a storage bucket, or attached to an account — there is no account on this page and nothing to sign in to. Both exist only for as long as the request takes.
What we do keep is anonymous and numeric: that a test started, that it finished, your filler count and rate, and any error, so we can tell whether the tool is actually working. No words, no audio, no name.
How it works
Three steps, thirty seconds
Tap and talk
Allow the microphone and talk for about 30 seconds. Say something you would say anyway — what you did this morning, how your week is going. Keep going until the bar fills.
Your clip is transcribed
The recording is sent to our speech-to-text provider and comes back as text. This is the step that leaves your device, and it is the only way to count words rather than sounds.
We count three words
Every um, uh and like in the transcript is counted, then divided by the time you spent speaking. You get the total, the rate, and where that sits on this page's scale.
What the rate means
The rate is your filler count divided by the time you were actually speaking, so a take with long pauses is not scored as though you had been talking through them. The cut points below are this page's own, chosen so that a 30-second take changes band on a whole number of fillers. They are not a population average — we have no reference distribution for filler rate, so this scale says where a take sits on this page's scale and nothing more.
| Band | Per minute | In a 30-second take | What it sounds like |
|---|---|---|---|
| Hardly any | Under 2 | 0 | Almost no fillers in that take. Whatever you were doing there — taking a beat of silence instead of a sound — is the thing to keep doing. |
| Light | 2 to 5 | 1–2 | A few fillers, spaced out. This is what unrehearsed speech usually sounds like, and it reads as natural rather than hesitant. |
| Noticeable | 6 to 11 | 3–5 | Enough that a listener would start to hear them. Fillers cluster where you are deciding what to say next, so this is usually about pacing rather than vocabulary. |
| Frequent | 12 or more | 6 or more | Fillers were carrying most of your pauses in this take. That is a habit rather than a trait, and it is one of the more trainable things about how you speak — a deliberate pause does the same job. |
This describes 30 seconds of one take, not how you always speak. Filler rate moves a lot with the situation: the same person lands in different bands explaining something familiar and thinking out loud about something new.
FAQ
Filler words, explained
What counts as a filler word here?
Three words: um, uh and like. Nothing else. Plenty of other things get called fillers — you know, so, right, actually, I mean — and they behave differently enough that folding them in would make the number harder to read rather than more complete. If this page ever counts more, it will say so here.
Do you count every “like” I say?
No, and this is the part worth understanding. “Like” is an ordinary word — “I like it”, “a day like today” — and counting those would hand you a number that means nothing. This page counts “like” only where it is standing in for a pause, and skips the uses where it is doing real work in the sentence. It is a rule of thumb rather than a grammarian, so it will miss some and over-count others. The um and uh figures are the more reliable half of your total.
How often do I say um — is 30 seconds enough to tell?
Enough for a reading, not enough for a verdict. Thirty seconds usually holds one to three fillers, and one either way moves your per-minute rate by about two. Run it two or three times on different things you might actually say, and the range you get back is more honest than any single take. If a take was too short to put a rate on, this page tells you instead of guessing.
Is my voice recorded or uploaded?
Yes — the clip is sent off your device, and that is the one thing this tool cannot avoid. Counting words means reading words, and reading words means speech-to-text, which runs on a provider's servers rather than in your browser. Your clip goes over an encrypted connection to this site and on to Deepgram, which returns the text. We do not store the audio and we do not store the transcript, and there is no account here to attach them to. Our other voice tools work differently: the Voice Hz Test and the monotone test measure sound rather than words, so they finish without sending a clip anywhere.
Are filler words actually a problem?
Not inherently. They buy you a moment to think, and every fluent speaker uses some. They start to cost you when they fill every pause, because a listener hears the searching instead of the point. The useful goal is not zero — it is being able to leave a pause silent when you want to.
Can I actually say fewer of them?
More readily than most things about how you speak. A filler fills a gap, so the training is mostly about getting comfortable with the gap: slow down slightly, finish the sentence, let the silence sit. Recording yourself is most of the work, because the pattern is nearly invisible from the inside — which is what this page is for.
Does this work in languages other than English?
No — this page is US English only. Filler words are language-specific (the German “äh”, the French “euh”, the Japanese “えーと”), and a counter built for one language reads the others wrong. Transcription is set to English, so a clip in another language comes back as noise or as nothing at all.
Keep going
You can hear the habit now. The rest of how you sound is measurable too.
Deepr gives you a daily voice score built on what it measures — how deep you sound, how steady your pitch is, how consistent your volume is — with exercises for the depth side of that. Counting filler words is this page's job, not the app's.
