An “um” before a sentence is easy to cut. An “um” that runs straight into the next word takes a closer listen. You can clean up a recording one word at a time in the transcript or ask an agent for a first pass and review the joins.
This guide is the 2026 workflow for cleaning up filler and silence without losing the rhythm that makes the clip feel human. I’ll show how to do it in ChatCut because that’s where I work, and I’ll cover Descript and Gling fairly because they invented this category and remain strong choices.
Why remove silences and filler words at all?
The math is simple: cleaner audio means higher completion rate, and completion rate is the dominant signal in every modern social and YouTube algorithm.
Concrete impact, from the audio-cleanup work I’ve watched across creator workflows:
- A 60-minute podcast tightened to 50 minutes by removing silences usually gets 12-18% higher completion rate
- A 5-minute YouTube tutorial with filler words removed (typically 30-90 seconds of “um”s and “uh”s in a 5-minute talk) consistently outperforms its uncleaned version on session watch time
- A vertical short clip cut from a podcast lands far better when the source has been tightened first; the 25-second window has no room for dead air
The audience effect is equally real. Silences and filler words make speakers sound less authoritative; the same content delivered tightly reads as more confident. Even when the audience can’t articulate why, they respond to the cleanup.
How do you remove filler words automatically?
The 2026 workflow uses text-based editing as the substrate. Instead of scrubbing the timeline looking for “um”s, you let the AI find them and you decide which to remove.
Start with a request that tells the ChatCut Agent which words to keep.
Remove clear hesitations such as “um” and “uh” from this video. Keep “like” and “you know” where they carry meaning, and preserve natural breaths.
For a manual cut, open Workspace → Transcript, select the unwanted “um” or “uh,” and press Backspace or Delete. ChatCut removes that word’s time range and closes the gap on the selected track. Listen across the join to make sure the next word still starts cleanly. Transcript and Captions
The same workflow in Descript looks similar; Descript pioneered the filler-word removal feature and it’s still excellent. Descript’s Remove Filler Words tool catches the standard set (“um”, “uh”, “like”, “you know”, “so”, “actually”) with one click. Gling AI does the same thing with a slightly different UI.
The differentiation between these tools in 2026 is mostly:
- What counts as a filler word. Some creators want “actually” and “so” preserved (they’re not really fillers in some contexts); others want everything cleaned. Most modern tools let you customize the filler list.
- How aggressive the trim is. Tight trims remove the filler plus the surrounding pause; loose trims remove only the filler word itself. ChatCut and Descript both let you tune this.
- Whether the AI removes only words or also non-word fillers. Breath pauses, lip smacks, mouth clicks are technically not filler words but listeners hear them as audio noise. Tools differ on whether they catch these by default.
For most workflows, the default settings work fine. The customization matters for high-stakes content where the speaker’s natural cadence is the brand.
How do you remove silences from a video?
Silence removal is mechanically similar but conceptually different from filler removal. A filler is a recognizable word the AI can match against a list. A silence is the absence of audio, defined by a threshold (anything quieter than X dB for longer than Y milliseconds).
The ChatCut workflow:
Remove all silences longer than 0.7 seconds from this video
In the prompt above, 0.7 seconds tells the Agent which gaps to cut. For a manual pass, Pauses in the Transcript panel sets the pause length you want to keep across the selected track. You can lengthen pauses again up to the time available in the source recording.
The threshold setting is the most important variable. Too loose and you don’t remove enough; too tight and you remove the natural pauses that make speech feel human. The right setting depends on the content:
- Conversational podcasts: 0.8-1.2 seconds. Preserve breathing room.
- Solo talking-head: 0.5-0.7 seconds. Tighter pacing without sounding rushed.
- Tutorial / explainer: 0.3-0.5 seconds. Maximum pacing for retention.
- Interview cuts for social: 0.3 seconds. Tight, social-first delivery.
If your recording has long breaks between takes, Clipchamp’s free Auto cut flags pauses longer than three seconds and lets you preview each one. For a half-second hesitation, use a manual cut. Microsoft’s silence-removal guide
For talking-head and interview workflows specifically, silence removal is usually the first cleanup step. Filler word removal comes second. Manual review (for any cut that took out something you wanted to keep) comes third.
Should you remove ALL pauses?
No. This is where automated cleanup gets a bad reputation, and where most beginners over-correct.
Pauses serve rhetorical purposes:
- The dramatic pause before a punch line. If the speaker pauses for 2 seconds before “and that’s why we lost the deal”, that pause is the joke.
- The thinking pause when the speaker is genuinely processing. Removing this makes the speaker sound robotic.
- The breath pause at the end of a sentence. Without it, sentences run into each other and the listener can’t follow the structure.
- The transition pause between topics. Helps the audience reset before the next idea.
The honest workflow: run automated cleanup at a moderate threshold (0.7 seconds), then watch the result with a finger on the pause button. Wherever the cleanup made the audio feel rushed or unnatural, restore the original pause manually. ChatCut’s text-based editor lets you do this without losing the rest of the cleanup.
A useful rule: aim for a cleaned version that’s 80-90% of the original duration, not 60-70%. The deepest cuts read as over-edited; the moderate ones read as professional.
Descript vs ChatCut vs Gling: how do they differ?
All three handle the same core job (text-based editing with automated silence and filler removal). The differentiation is in everything else.
Descript. Invented this category. The most mature single-purpose tool for podcast editing and talking-head cleanup. Strongest standalone feature set: filler removal, Studio Sound (audio enhancement), Overdub for voice replacement (with constraints), and Underlord (their conversational editing assistant) for higher-level edits. Descript Free includes 60 minutes of processed media per editor each month and exports video at up to 720p with a watermark. Descript plan comparison
ChatCut. Combines text-based editing with a fuller editing surface (motion graphics generation, AI video generation via Seedance 2.0, multi-track timeline). The wedge: one prompt drives multi-step edits (“remove silences AND add captions AND export 9:16”). Browser-based, no install. Gling AI. Lightweight, focused tool. Specifically built for the “long-form recording → tightened video” workflow. Less full-featured than the other two but quick to learn and reliable on the narrow job. Gling Free includes one hour of AI-edited media per month and watermarks MP4 exports. Gling plans
The honest meta-recommendation: pick Descript if you’re a podcast-only producer who values the most mature filler removal and Studio Sound; pick ChatCut if you’re producing video that mixes talking-head with other formats (motion graphics, AI B-roll, social repurposing); pick Gling if you want the simplest possible tool focused only on this one job.
For audio that needs more than silence and filler cleanup (background noise, mic bleed, room reverb), AI noise removal handles a different layer of the problem and works in combination with silence/filler removal.
Start with noise removal if the background masks the speech. Shorten long gaps, remove unwanted fillers, then listen through the joins for clipped syllables and missing breaths.
FAQ
Is ChatCut free for this?
Yes. Transcription and manual cuts in the Transcript panel use no ChatCut credits on Free. A local export in the browser or ChatCut Desktop also uses no credits. For prompt-based cleanup, the built-in Agent uses ChatCut credits. A connected Claude Code or Codex agent uses your host subscription or API allowance for the same edits. Free and Paid Plans · Credits
Can these tools remove “um” but keep “you know” if I want?
Yes. In ChatCut, ask for “um” and “uh” to be removed and for “you know” to stay where it carries meaning. For a single ambiguous word, listen to the sentence and select just the word you want to cut in the transcript. Talking Head Edit
What happens if the AI removes a word I wanted to keep?
ChatCut and Descript both let you undo any specific cut. The text-based editor shows the removed words struck through; you can restore them individually. Most workflows go: run the cleanup, watch the result, restore the 5-10 words you actually wanted.
Does silence removal affect the audio quality of the surrounding speech?
Listen to a second on either side of the cut. A clipped consonant or a missing breath can make two intact words sound spliced together. Restore a few frames and play the join again.
How long does it take to clean up a 60-minute podcast?
Allow time for transcription and a listening pass. A dense interview with overlapping speakers needs more review than a solo recording with long gaps between takes.
Will this work on languages other than English?
For silence removal, yes; silence is silence in any language. For filler word removal, the tool needs a filler-word list for the target language. ChatCut and Descript both support filler removal in major languages including Chinese, Spanish, French, and German. Less common languages are hit-or-miss.
Try the cleanup workflow
Open ChatCut, upload a recording you’ve been meaning to clean up, and try:
Remove all filler words and silences longer than 0.7 seconds from this video
Play the tightened timeline with sound on. Restore any pause that carried a punch line or gave the listener time to follow a change of subject.