mimicTo remove silences, cut every stretch where the waveform goes flat between words and sentences, and keep a couple of frames either side of each word so nothing sounds clipped. Do it by hand on the waveform, or let a tool detect pauses below a loudness threshold for longer than a minimum duration.
For short-form talking video, treat pauses over about 0.3–0.5 seconds as dead air. Then remove ums and false starts, which silence detection can’t see.
By the numbers
- −60 dB / 2 sFFmpeg's default silence threshold and minimum duration: far too strict for speechFFmpeg, 2026
- 3 secTikTok's window for introducing what your video is about. Dead air spends itTikTok Ads Help Center, 2025
- 10–30 sLength Google suggests for action-focused vertical Shorts adsGoogle Ads Help, 2026
Why dead air kills short-form
On a feed, a pause is an exit. TikTok tells advertisers to introduce the point in the first three seconds and land the hook in the first six (TikTok Ads Help Center, 2025). A breath, an “okay, so”, and a half-second of looking at the screen can eat most of that before you’ve said anything.
Removing silences also tightens the whole video. If a brand asked for 30 seconds, every second of dead air is a second of product you didn’t show. Google’s own guidance for action-focused Shorts ads is 10–30 seconds (Google Ads Help, 2026), which leaves no room for pauses.
What to cut, and what to keep
| Cut | Keep |
|---|---|
| Dead air at the start and end of every take | A beat after a punchline or a reveal |
| Pauses between sentences longer than ~0.3–0.5 s | Breaths inside a sentence |
| Ums, uhs, 'so', 'like', 'okay' | Natural rhythm in a story |
| False starts and repeated lines | The single best version of each line |
| Looking down at notes, reaching for the phone | Pauses that show the product working (a pour, a click) |
The goal isn’t zero pauses. It’s no pause that doesn’t do a job.
How to remove silences by hand
Any editor with a visible audio waveform works: CapCut, Premiere, DaVinci Resolve, Final Cut, even most phone editors.
- Make the waveform big. Zoom the timeline in and enlarge the audio track. Speech shows as peaks; silence as a flat line.
- Trim the head and tail of each take first: everything before your first word and after your last.
- Work left to right. At each flat gap, split just before the next word starts and just after the previous one ends. Delete the gap and close it up (ripple delete).
- Leave handles. Two or three frames of room either side of each word. Consonants like “s” and “t” trail off quietly and are easy to clip.
- Remove the stumbles. Listen at normal speed and delete false starts and repeated lines, keeping the best take of each.
- Play it back at normal speed, with headphones. Fix any cut that sounds clipped or rushed.
It’s slow, but you control every cut.
How automatic silence removal works
Every auto tool, whatever it’s called, works on two numbers:
- Threshold (dB): how quiet audio has to be to count as silence.
- Minimum duration: how long it has to stay that quiet before it’s cut.
The open-source FFmpeg silencedetect filter is a good reference: it flags audio at or below a noise tolerance for at least a minimum duration, with defaults of −60 dB and 2 seconds (FFmpeg, 2026). Those defaults are built for finding long gaps, not tightening speech. A talking video recorded in a normal room needs a higher threshold and a much shorter minimum.
| Setting | Start here | If it goes wrong |
|---|---|---|
| Threshold | Just above the noise floor between words (often around −35 to −45 dB in a quiet room) | Cutting quiet words: lower it. Missing pauses: raise it |
| Minimum pause | 0.3–0.5 s for short-form, 0.6–1 s for longer, calmer videos | Too choppy: make it longer |
| Padding / handle | ~0.05–0.1 s each side | Clipped words: add more |
These are starting points, not rules. Read your own waveform: the noise floor in a quiet bedroom and a busy kitchen can be very different, and the right threshold follows the room.
Silence detection only finds quiet. It won’t touch “um”, a cough or a false start, because those are loud. For those you need a transcript-based tool or your ears. Mimic handles both together: it cuts the pauses and the false starts, then captions what’s left word by word.
Fixing choppy audio after cutting
- Clipped words: extend the handle on the affected cut by a few frames.
- Breathless delivery: you removed breaths inside sentences. Put them back; remove only breaths between sentences.
- Clicks at cuts: add a very short audio crossfade (a few milliseconds up to a couple of frames) on each cut.
- Room tone jumps: background hiss changes between takes. Even it out with light noise reduction across the whole clip, or a quiet music bed.
- Volume jumps: different takes were recorded louder or quieter. Level the voice so it sits at one consistent loudness throughout.
- Visual jumps: every cut is now a jump cut. Hide the distracting ones with punch-ins or B-roll; see the jump cuts guide.
Before and after
[1.2 s silence] “Okay so… [0.8 s] this is the, um, the sunscreen I’ve been using [0.6 s] — wait. [1.5 s] This is the sunscreen I’ve been using every day, and it doesn’t leave a white cast.” [1 s]
“This is the sunscreen I’ve been using every day, and it doesn’t leave a white cast.”
Same sentence, same take. Everything removed was either silence (head, tail, gaps) or a stumble (“okay so”, “um”, the false start). Then punch in on “doesn’t leave a white cast” and caption it.
Record so silences are easy to cut
- Leave a clean pause between lines. A full second of quiet is easier to cut than a run-on “and-um-so”.
- Re-say the whole line after a stumble, not just the word you missed.
- Quiet room, mic close. The lower the noise floor, the cleaner any threshold works. See how to film UGC on iPhone.
- Read from a teleprompter placed under the lens to cut down on look-away pauses.
Questions
How do I remove silence from a video automatically?
Use a silence-removal or 'delete pauses' feature in your editor, or a tool built for it. It finds stretches below a loudness threshold that last longer than a minimum duration and cuts them. Set the threshold just above your room noise, the minimum pause to about 0.3–0.5 seconds for short-form, and add a small handle so words aren't clipped. Then watch the result.
What threshold should I use to detect silence?
It depends on your recording. Look at the waveform: the threshold should sit just above the noise floor between words. Generic defaults are often far too strict for speech; FFmpeg's silencedetect, for example, defaults to -60 dB and 2 seconds, which misses most pauses in a talking video.
Why does my audio sound choppy after removing silences?
Usually the cuts are too tight (word starts and ends clipped), breaths inside sentences were removed, or the room tone changes at each cut. Add two or three frames of handle, keep breaths within a sentence, and put a very short crossfade on each audio cut.
Should I remove all pauses from a video?
No. Remove pauses between sentences and the dead air before and after takes. Keep short pauses that do a job: after a punchline, before a reveal, or where a breath makes the line sound human.
How do I remove 'ums' and false starts?
Silence detection won't catch them, because they aren't quiet. Use a transcript-based editor that marks filler words, or cut them by hand. Mimic removes pauses and false starts together, so you keep only the best version of each line.
Does removing silences create jump cuts?
Yes. Every cut in a continuous shot is a jump cut. That's normal on short-form; smooth the noticeable ones with a punch-in or B-roll.
Sources
- FFmpeg Filters Documentation: silencedetect, FFmpeg (2026). Silence = volume at or below a noise tolerance (default -60 dB) for at least a minimum duration (default 2 seconds)
- Creative best practices for performance ads, TikTok Ads Help Center (2025). Introduce the proposition in the first 3 seconds; hook in the first 6
- Your guide to YouTube Shorts ads, Google Ads Help (2026). 10–30 second vertical videos recommended for action-focused campaigns; only the first 60 seconds of longer ads play in the Shorts feed