How to remove filler words from a video (ums, uhs and likes)

Fillers are the cheapest thing to fix in a talking video and the most noticeable when you don’t. Nobody minds the odd “um”. Fifteen in two minutes makes a clear explanation sound unsure. Here’s how to get rid of them, by hand, with tools, and at the source.

What counts as a filler

There are two kinds, and they need different treatment.

• Sounds: um, uh, er, erm, ah, hmm, mm. These never carry meaning, so they’re safe to remove in bulk. Every automatic tool goes after these.
• Words: like, you know, so, basically, actually, I mean, right. Sometimes they’re filler, sometimes they’re doing a job. “So that’s why the build fails” loses something if you take the “so” out. “I like this approach” falls apart without “like”. These need a human.

The aim isn’t zero fillers. A video with every hesitation removed sounds like it was assembled, not spoken. I cut every um and uh, then only the word fillers that sit in a run (“so, like, basically”) or start a sentence for no reason.

Doing it by hand in a normal editor

If your editor has no transcript, you can still do this quickly once you know what an um looks like. On the waveform it’s a short, smooth, even blob with no consonant spikes, usually after a small gap. After the first few you’ll spot them without listening.

1. Play at normal speed and tap M to drop a marker on each filler (in Final Cut Pro, Premiere and Resolve). Don’t stop to cut yet: one listening pass is faster than stopping and starting.
2. Go back to each marker. Zoom right in on the timeline so a word is a few centimetres wide.
3. Split just before the filler and just after it. That’s ⌘B in Final Cut Pro, iMovie and Resolve, and ⌘K (Add Edit) in Premiere.
4. Select the piece and ripple delete it so the gap closes: ⇧⌫ in Resolve, Edit > Ripple Delete in Premiere. In Final Cut Pro and iMovie a plain ⌫ closes the gap.
5. Play across the cut from a second before. If you hear a click or a clipped consonant, move the cut a frame or two.

Final Cut Pro 12 (January 2026) added Transcript Search on Apple silicon Macs, so you can search for “um” and jump between hits instead of listening for them. It finds them, it doesn’t cut them, but it saves most of step 1.

It works, but it’s slow: a ten-minute take with a filler every few sentences is a lot of splits. That’s the main reason people reach for a tool.

Making the cuts not obvious

Removing a filler leaves two problems: a jump in the picture and an unnatural join in the sound. The sound matters more. People forgive a small visual jump far more readily than a voice that changes pitch mid-breath.

Cut on breaths

Most ums sit next to a breath: breath, um, sentence. Cut the um and keep the breath, so the next sentence starts the way a sentence naturally starts. If you cut the breath too, the sentence starts abruptly and sounds stitched. If there are two breaths, keep one.

Leave a little air

Don’t cut tight against the next word. Leave about two frames (around 80 ms) before the first consonant, or you’ll clip the start of it, and “table” becomes “able”. A one- or two-frame audio crossfade across the join stops clicks.

Fill gaps with room tone

Every room hums a little: a fan, traffic, the fridge. If you cut a filler and replace it with true silence, the hum drops out and the ear notices. Record ten seconds of silence in the same spot before or after you talk, and lay a piece of it under any gap. If you forgot, copy a quiet stretch from elsewhere in the take.

Use a J-cut

In a J-cut the sound of the next shot starts slightly before its picture. Because the ear hears a continuous voice, the picture jump a few frames later doesn’t register as a cut. In Final Cut Pro, select the clips, press ⌃S to expand audio, then drag the video edge so the audio overlaps (Apple’s steps are in the split edits guide). Premiere and Resolve do the same by unlinking or trimming audio and video separately.

Punch in on the webcam

On a talking head, scale the clip after the cut up by 10 to 15 percent. The jump now looks like a deliberate camera change. Alternate between two framings and nobody counts the cuts. On a screen recording with a small webcam bubble, the jump is barely visible at all, which is why screen videos can take far more cuts than a full-face video.

Resolve’s Smooth Cut transition and Premiere’s Morph Cut try to morph across a jump cut automatically. They work when you barely moved and the background is plain, and they warp your face when you didn’t.

Removing fillers automatically

Every automatic tool does the same thing underneath: transcribe the audio, time each word, flag the fillers, cut them. The differences are in what gets flagged, how carefully the cut lands and where your video goes. This is what each one does at the time of writing (October 2026), from their own documentation.

Tool	What it flags	Notes
Premiere Pro	Fillers such as um and ah	Text panel: filter the transcript to show fillers, review, delete.
Descript	Common fillers like um and uh	English transcripts only. Uses AI credits. Has an “Avoid harsh cuts” option.
Camtasia Audiate	um, uh, like, you know	Delete all at once or review each one. Edits carry over to Camtasia.
Riverside	um, so, uh	Choose Smart, Cut or Mute.
Loom	um and ah, or a broader setting with rambles and repeats	Business + AI plan.
Auphonic	um, uh and their German, French and Spanish forms	Audio only: automatic cutting isn’t available for video files.
Blazing Video	um, umm, uh, uhh, uhm, er, erm, ah, hmm, mmm, mm	Mac only. Runs on the Mac, nothing uploaded for transcription.

DaVinci Resolve has no filler detector. The paid Studio version transcribes clips and lets you delete words from the transcript, so you can find and cut them by eye, but nothing flags them for you.

Whichever you use, check the result. Loom’s own help page says it can’t catch every filler, and transcripts sometimes write a real word as “uh” or swallow a filler into the next word. Listen to the edited version once at 1.5× before you send it.

Which fillers tools catch, and which they miss

Sounds are the easy part. They look the same in every transcript and nobody says “um” on purpose, so the tools above all get most of them.

“Like”, “you know” and “so” are the hard part. Auphonic leaves out “like”, “well” and “you know” deliberately, because they often carry meaning. Camtasia flags “like” and “you know” but tells you to check each one. Tools that remove word fillers in bulk are the ones that produce sentences that don’t quite parse.

My approach: let the tool take the sounds in one pass, then search the transcript for “like”, “you know” and “basically” and decide each one. In a ten-minute video that’s usually a dozen decisions, two minutes of work.

Tools also miss the fillers that aren’t words: the sentence you started, abandoned and restarted. Those are often worse than any um, and a filler button won’t touch them. Look for repeated phrases in the transcript and keep the last attempt.

Doing it in Blazing Video

Blazing Video is a Mac app (macOS 15 or later, free during the private beta) that records your screen and webcam and transcribes on the Mac about half a second after you stop. Every word is then aligned to the audio, so cuts land between words rather than through them.

1. Press ⌥⇧R to record and again to stop.
2. In the AI tools panel, choose Cut fillers. The ums and uhs are tinted in the transcript first, so you can see what will go, then cut in one undo step.
3. For “like” and “you know”, select the words in the transcript and press ⌫. They’re struck through and the video skips them. ⌫ again or ⌘Z brings them back.
4. Remove retakes handles the restarted sentences: it finds lines you said twice and keeps the last attempt.
5. Cut pauses shortens gaps longer than 0.4 s down to 0.15 s by default. I’d keep it around there; shorter starts to sound breathless.

If you’ve already sent the share link, press Update Link and the same link plays the cleaned-up cut, captions and transcript included. There’s a longer walkthrough in how to edit a screen recording by deleting words, and if you use Claude you can say “cut the ums and tighten the pauses” and let Claude do it.

It isn’t the right tool for everything. It’s Mac only, there’s no Windows or web version, and if you’re editing a podcast audio file, an audio-first tool like Auphonic fits better.

Saying fewer in the first place

An um is your brain holding the floor while it finds the next sentence. You can’t stop the search, but you can change what fills it.

• Pause silently instead. A silent half-second sounds thoughtful on video, and it’s trivial to tighten later. An um has to be cut. Practise by closing your mouth when you feel one coming.
• Know the next sentence before you start it. Fillers cluster at the start of sentences and at transitions. Three bullet points on a sticky note beside the camera cut most of them.
• Record in short sections. Fillers multiply when you’re trying to hold a ten-minute structure in your head. Record one point, stop, glance at your notes, record the next.
• Restart instead of recovering. If you fluff a sentence, pause and say the whole sentence again. One clean retake is easier to cut than a sentence patched with “uh, I mean”.
• Watch one back. Count your fillers in a single video. Most people have one or two habits, often “so” at the start of a sentence, and knowing yours is half the fix.

FAQ

Can I remove filler words from a video for free?

Yes. Any editor can do it by hand: iMovie and DaVinci Resolve’s free version both let you split around each filler with ⌘B and delete it. It just takes longer. Blazing Video’s Cut fillers is free during its beta, on a Mac.

Does removing filler words make a video sound choppy?

It can, if cuts clip the next word, remove breaths or leave dead silence. Leave a couple of frames of air before the next word, keep one breath, and lay room tone under gaps. Cuts made from word-level timings tend to land more cleanly than cuts made by silence detection.

Why didn’t the tool remove “like” or “you know”?

Because those are often real words, most tools leave them out or ask you to check each one. Search the transcript for them and delete the ones that are filler yourself.

Should I remove every um?

Remove the sounds, keep the occasional natural pause. A video with every hesitation and breath stripped out sounds machine-made. Aim for clean, not perfect.