How to cut pauses and “um”s from a video automatically
Pauses and “um”s are the main reason a talking video feels slow. Cutting them by hand takes ages: a minute of footage means dozens of cuts. Here’s how automation does it, why simply “cutting silence” sounds bad, and what VARIA does differently.
- Cuts pauses, “um”s and “hmm”s, filler words, repeats and sentences started over
- Numbers, negations and meaningful words are protected
- Cuts only in silence — no clipped words
- Street noise is reduced by a neural network
- “Smart cut” can be turned off
Why “cutting silence” isn’t enough
The simplest way is to find spots where the volume drops below a threshold and cut them. That’s how many automatic “pause removers” work. There are three problems:
- “Um” isn’t silence. A drawn-out sound is loud, the threshold doesn’t catch it, and it stays.
- Clipped word endings. The last sound of a word is often quiet — it gets cut along with the pause, and the speech sounds “swallowed”.
- Choppy rhythm. If you cut everything, the short breaths between sentences disappear too — the speech becomes robotic.
How VARIA does it
The edit looks at the text and the sound at the same time:
- Timing of every word. Speech recognition returns not just text but the start and end of every word.
- Volume every 10 ms. It’s used to refine word boundaries: the cut goes into real silence, not into the middle of a fading sound.
- Drawn-out sounds. “Um” and “hmm” are recognized by sound — loud but almost unchanging — and cut, even if the recognizer “hid” them inside a neighboring word.
- Fillers and parentheticals. “Well”, “like”, “I mean”, “kind of”, “actually” used as fillers are cut; if you say “I mean” out of habit in every sentence, they’re all removed. But if a filler is said in one breath with the words around it, it stays: a cut in the middle of continuous speech sounds like a jump.
- Started over. If you stumbled and repeated a sentence, the second, successful take stays. A rephrasing of the same sentence a couple of seconds later is caught too.
- Off-camera remarks. “Action”, “Stop”, “One more time”, “What else should I say?” at the end and “bye-bye” after a finished thought are cut.
What’s protected from cutting
Numbers, negations and conditions are never cut — even if they look like a repeat. “Third-to-fourth degree” won’t turn into “third”, and “don’t do it” won’t become “do it”. If almost half of the video is up for cutting, the edit switches to a gentle mode: in a dialogue with pauses between lines, it’s better to keep some air than to lose meaning.
How tightly pauses are cut
Pauses between sentences are squeezed to about a tenth of a second — the pace at which dynamic talking videos are edited today. A one-minute video usually loses 15–40% of its length, depending on how you speak. The end of the video is handled separately: after the last word, half a second stays, and if the camera jerked at the end (you reached to press “stop”), the video ends before the jerk.
How long this takes by hand
In a mobile editor, a pause is cut like this: find it on the track, split at the start, split at the end, delete the clip, check by ear. That’s 15–30 seconds per cut if you do it carefully. A minute of live speech usually has 15–30 spots worth cutting — that’s 5–15 minutes on pauses alone, without captions or a title. Plus, after every cut, the captions shift if you made them in advance.
Automation does the same in the time it takes to recognize the speech — seconds per minute of footage — and doesn’t get tired by the twentieth video.
How to control it
- “Smart cut” — one switch: on — pauses and fillers are cut; off — length and words stay as they are.
- Notes in your own words — “remove the first sentence”, “keep the pause before the conclusion”.
- Editor — any piece can be restored or cut manually; captions and text overlays shift along with the video.
Sound after cutting
Cuts are smoothed with short fades, the whole video’s volume is evened out, and if the recording is noisy, the background is reduced by a neural network — gently, without “vacuum” silence in the pauses. Picture and sound are cut in sync, so lips don’t drift from the voice even after dozens of cuts.
Pricing
Auto-edit is paid in packs: 5 edits — $7.13; 10 edits — $13.18; 25 edits — $30.10; 50 edits — $54.28. One edit is one finished video from a clip up to 3 minutes. Your first video can be a trial — with a watermark. Reposts are separate, from $6.05 for 30 reposts; 2 reposts free to start. Packs never expire.
FAQ
Can I remove only pauses, without the rest of the design?
The lightest option is the “Captions only” preset: clean speech and text, no title, graphics or music. Pauses and fillers are cut with “Smart cut” on in any style.
Won’t it cut a word I need?
Numbers, negations and conditions are protected. If something is cut by mistake, you can bring it back in the editor or with a request in the notes.
Does it work with a dialogue between two people?
Yes, but pauses between lines are cut more gently: if cleanup would remove more than 45% of the video, the edit switches to a gentle mode automatically.
What if there’s street noise in the recording?
If the speech noticeably drowns in noise, the background is reduced by a neural network. Quiet recordings are left alone so the voice isn’t damaged.
How much length is usually cut?
Depends on how you speak: most often 15–40% of the video. Recordings with long hesitations get shortened more.