Workflow guide
Tighter pacing without the jumpy, breathless edit
Cutting dead air is the fastest way to raise retention, but aggressive silence removal makes speakers sound anxious and cuts betray every edit. The craft is knowing which pauses carry meaning and which ones waste time. This guide shows a threshold-and-padding method that works in any editor.
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
Detect pauses below roughly minus 40 dB lasting over half a second, keep 150 to 250 milliseconds of padding on each cut, preserve pauses before punchlines and transitions, smooth the joins with short audio crossfades, then review at full speed for natural breathing.
Pace without panic
Dead air is gone but the speaker still sounds like a calm human, not a chipmunk.
Invisible joins
No clicks, room-tone jumps, or half-swallowed first syllables at cut points.
Meaning survives
Dramatic and comic pauses stay; only empty waiting time is removed.
Set detection so it respects speech
Silence detectors watch for audio dropping below a loudness threshold for a minimum duration. Start near −40 dB with a minimum pause of 0.4–0.6 seconds for conversational speech, then preview before committing. A threshold set too hot eats soft consonants and breathy line endings; one set too cold leaves every mid-sentence hesitation in place.
The padding around each cut matters more than the threshold itself. Keep 150–250 ms of room tone on either side so entrances do not clip and exits do not slam. For dramatic content, exempt pauses before key lines entirely — mark protected ranges first, then run detection on the rest. In Mellius you can describe this directly, such as asking to tighten pauses but keep dramatic beats, and adjust the result on the timeline.
- Start at −40 dB, 0.5 s minimum, 200 ms padding.
- Protect pauses before punchlines, reveals, and topic changes.
- Never cut inside a word's reverb tail or breath.
- Check soft-spoken guests separately — one preset rarely fits all mics.
Hide the joins and check the rhythm
Cuts in room tone are what make silence removal audible. Add 10–30 ms constant-power audio crossfades at each join to kill clicks, and where the background hiss audibly steps, bridge the gap with a bed of matching room tone under the cut. Video can hide behind a B-roll cutaway or a two-frame dissolve if a jump cut lands mid-gesture, but audio smoothness is non-negotiable.
Review the whole edit at 1x speed, not scrubbed, and listen for three failure modes: stacked cuts that remove all breathing room, music or ambience that pumps unnaturally, and sentences that now collide. A useful target is removing 10–25% of runtime for interviews; beyond 30%, reconsider whether the script itself needs cutting rather than the pauses.
- Add 10–30 ms audio crossfades at every silence join.
- Cover unavoidable jump cuts with B-roll or punch-ins.
- Listen once at full speed before exporting.
- If cuts exceed 30% of runtime, shorten the script instead.
Compare the workflows
| Method | Best fit | Trade-off |
|---|---|---|
| Manual razor cutting | Short videos and maximum control | Too slow for long interviews |
| Auto silence detection | Podcasts, interviews, and courses | Needs threshold tuning and a review pass |
| Text-based deletion | Dialogue-heavy edits with transcripts | Can miss visual-only dead moments |
A practical sequence
Put it into practice
- 01
Protect meaningful pauses
Mark dramatic beats, punchlines, and transitions to exempt before any auto-cutting.
- 02
Detect and pad
Run detection near −40 dB and 0.5 seconds, keeping 150–250 ms of room tone per join.
- 03
Smooth and review
Add short crossfades, cover jump cuts, and listen end to end at normal speed.
Frequently asked questions
Why does my video sound choppy after removing silences?
Choppiness usually comes from zero padding or missing crossfades, so entrances clip and room tone jumps. Re-add 150–250 ms of padding and 10–30 ms audio crossfades at each join. If it still feels rushed, you cut meaningful breathing pauses — restore the ones before emphasis and transitions.
What silence threshold should I use?
Start around −40 dB with a 0.4–0.6 second minimum for normal speech, then adjust per mic. Noisy rooms need a hotter threshold plus noise cleanup first, or the detector treats hiss as speech. Always preview on the quietest speaker before applying to the full timeline.
Should I remove all pauses from a talking-head video?
No. Remove empty waiting time but keep pauses that signal thinking, emphasis, or transitions — they help comprehension and make the speaker feel human. A common target is cutting 10–25% of interview runtime through pauses, then tightening further with script edits if needed.
Continue reading
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