Workflow guide
Cut the ums without making speech sound robotic
Filler words make even strong takes feel hesitant, but deleting every um usually creates jumpy, breathless dialogue. The goal is clean-read speech: hesitations gone, rhythm intact. This guide shows how to find fillers, tighten pauses, and smooth the joins.
From the Mellius product team · Updated September 28, 2026
Conspiracy theorists, you may hate this
AI Director
›
Just type what you need
- ›Remove all silent parts
- ›Add bold kinetic captions
- ›Only have Bill Ackman speaking
- ›Cut to highlights only
- ›Add cinematic transitions
- ›Match cuts to the beat
The short answer
Transcribe the video and flag ums, uhs, likes, you-knows, false starts, and long pauses. Delete fillers at zero crossings with two to four frames of room tone kept, tighten gaps between sentences to two to six frames, then listen at full speed and restore any cut that sounds rushed or breathless.
Speech still breathes
Cleaned dialogue keeps natural breaths and pacing instead of collapsing into a rushed wall of words.
Cuts are inaudible
Every join lands on a zero crossing or room tone so there are no clicks, pops, or ambience jumps.
Meaning is preserved
False starts and repetitions go, but emphasis, qualifiers, and personality stay in the performance.
Find fillers with a transcript, not by scrubbing
Scrubbing a timeline for ums is slow and misses half of them. Generate a transcript first — in Premiere that means Text-Based Editing, in DaVinci Resolve the Transcribe Audio button in the Media Pool, and in Mellius you can ask for a transcript and then ask to flag filler words conversationally. Search the text for um, uh, like, you know, I mean, and repeated sentence openers, then review each hit against the audio before cutting.
Separate true fillers from talking style. A thoughtful pause before a key point adds weight; a stammered restart of a sentence does not. Common pitfall: bulk-deleting every flagged word, which removes deliberate emphasis and leaves the speaker gasping. Keep fillers that buy a genuine beat or signal uncertainty the story needs, and cut the habitual ones that cluster at sentence starts.
- Flag ums, uhs, likes, false starts, and repeated openers in the transcript.
- Listen to each hit in context before deleting — keep deliberate beats.
- Handle stacked fillers (um-uh-like) as one cut, not three micro-cuts.
Cut and smooth so joins disappear
Cut on zero crossings with a two to four millisecond crossfade or a two to four frame room-tone bridge so the ambience does not jump. Tighten dead air between sentences to roughly two to six frames; anything shorter feels gasped, anything over half a second of silence reads as a mistake in fast content. When a cut removes a breath mid-phrase, paste room tone from elsewhere in the recording under the join rather than leaving a vacuum.
Always do a full-speed listening pass on headphones after cleanup, because waveform editing hides rushed rhythm. If the speaker sounds anxious or unnaturally fast, restore a few pauses — removing seventy percent of fillers usually sounds better than removing all of them. In Mellius the cleanup lands on the editable timeline, so you can ask to restore a pause or re-tighten a section and undo if the new version feels worse.
- Use tiny crossfades or room tone under every join.
- Target seventy percent removal first; polish only if needed.
- Check breaths — never stack two sentence cuts without air.
Compare the workflows
| Method | Best fit | Trade-off |
|---|---|---|
| Manual transcript cleanup | Maximum control over tone and rhythm | Slow on videos longer than ten minutes |
| Auto filler detection | Fast first pass on talking-head footage | Needs a human review pass to avoid robotic pacing |
| Clean-read rewrite with captions | Social cuts where reading speed matters | Not suitable for legal or verbatim deliverables |
A practical sequence
Put it into practice
- 01
Transcribe and flag
Generate a transcript, search for filler words and long pauses, and mark candidates without cutting yet.
- 02
Cut with room tone
Delete fillers at zero crossings, bridge joins with room tone, and tighten gaps to a few frames.
- 03
Listen at full speed
Play the whole section back, restore any cut that sounds rushed, and check joins on headphones.
Frequently asked questions
Should I remove every filler word from a video?
No. Removing every um and pause usually makes the speaker sound anxious and breathless. Aim to remove the habitual fillers that add nothing, and keep pauses that create emphasis or let an idea land. A seventy-percent cleanup sounds more natural than a total purge.
How do I avoid clicks when cutting filler words?
Cut at zero crossings where the waveform crosses the center line, and add a two to four millisecond crossfade on each join. For noisy rooms, lay a bed of room tone recorded from a silent stretch under the edit. Always check the result on headphones, where clicks are most obvious.
Is it better to edit fillers from text or the timeline?
Text-first editing is faster for finding fillers because you can search the words directly. Timeline editing gives finer control over breaths, overlaps, and ambience. The strongest workflow is both: flag candidates in the transcript, then fine-tune the joins on the timeline.
Continue reading
Primary sources
Product capabilities change. These official references support the comparisons above.
Describe the edit. Keep the timeline.
Mellius turns conversational direction into cuts, captions, graphics, motion, color, and sound you can still inspect and change.
Start editing free