Step one: find the quiet parts
ffmpeg doesn’t cut anything at first. You ask it to listen and report where the quiet stretches are:
ffmpeg -i talk.mp4 -af silencedetect=noise=-30dB:d=0.5 -f null -Anything quieter than −30 decibels for at least half a second counts as silence. ffmpeg prints a list of silence_start and silence_end times. That list is the whole result.
Step two: cut between the gaps
Now you keep everything outside those times. With the trim and atrim filters you cut each kept piece, then join the pieces back together.
There’s a shortcut that catches a lot of people out: the silenceremove filter. It works on audio only. Run it on a video and the sound gets shorter while the picture stays the same length, so the two drift apart. Video and audio have to be cut at the same moments.
Why the result sounds choppy
If you cut exactly at the edge of each gap, you clip breaths and the ends of words, and the speaker sounds like a robot. The fix is to leave a little air. The project we looked at trims long gaps down to 0.3 seconds instead of removing them.
How you cut matters too. Copying the video stream without re-encoding is fast, but it can only cut at keyframes, so the cuts land in slightly wrong places. Re-encoding is slower and exact.
Threshold problems
The right threshold changes with every recording. In a room with a fan or street noise, the background is louder than −30 dB, so ffmpeg finds no silence at all. Set it too high and soft words get cut. Gaps that matter, like the pause before a punchline, look identical to dead air.
What silence detection can’t hear
An “um” isn’t silence. Neither is a restart where you say a sentence again, or a take you repeated three times. To catch those, you need to know what was said.
That’s why newer tools add a transcript with a timestamp for every word. Then “cut the ums and keep my last take” is something you can compute, and it’s what Cutleap works from.
“Cut every pause longer than half a second, but keep the one before the punchline.”
“Cut the ums and restarts.”
“Keep only my last full take of each line.”
Want a fuller request? Copy the prompt Turn one talking clip into five short videos.
Let Cutleap do the editing
Upload your clips, say what you want, get finished videos back. 7-day free trial, cancel anytime.
Get early accessQuestions
Does ffmpeg remove ums?
No. ffmpeg only hears loudness, so it finds silence. Ums, repeats and restarts need a transcript.
Is ffmpeg free?
Yes, it’s open source, and a lot of video software is built on it. You run it from the command line.
Do I need to use the command line with Cutleap?
No. You upload your clip and say what you want.