# Remove silences and filler words automatically

Canonical URL: https://getvidiyo.app/features/silence-and-filler-word-removal
Last updated: 2026-10-08

> TL;DR: Type "cut the ums and long pauses" and Vidiyo removes them. It measures where your audio actually goes quiet, cuts on word boundaries from a local transcript, and keeps a little breathing room around speech. Repeated takes and false starts can go too. Every cut is on the timeline, and one undo restores them all.

Removing silences in Vidiyo is one request, but it isn't a blunt volume threshold. The assistant combines two kinds of evidence: where the audio is actually quiet, and where each word starts and ends in the transcript. It cuts only where both agree, so you lose the dead air without losing the ends of your words.

## How do I remove silences and ums from a video?

Import your talking-head clip and type the request in the AI panel. Vidiyo transcribes the audio on your Mac, finds the pauses and fillers, and cuts them on the timeline. You'll see a short preview under the reply and a list of what changed.

Things you can type:

- "Cut the ums and long pauses."
- "Tighten every pause longer than half a second."
- "Keep only the best take of each line."
- "Remove the false start at the beginning."
- "Start on the strongest line."
- "Punch in on alternate cuts so the jumps don't show."

If you want to see the plan first, switch the composer to plan mode. The assistant lists the cuts it intends to make and what it assumed, and waits for you to approve.

## How does automatic silence removal work?

When you import a clip, Vidiyo analyses the audio for quiet stretches: places where the level stays below -35 dB for at least 0.35 seconds. Those are pause candidates. The transcript adds word boundaries from Whisper's timing anchors, so Vidiyo knows exactly where the last word before a pause ends.

| Step | What happens |
| --- | --- |
| Measure | Quiet stretches are found in the audio on import |
| Align | Each transcript word gets a start and end time |
| Confirm | A pause counts only when the audio and the words agree |
| Cut | Confirmed pauses are removed, keeping padding around speech |
| Check | Quality checks flag any [dead air](https://getvidiyo.app/glossary/dead-air) or raw jump cuts left |

When the measurement and the word timing disagree, for example a word whose timing overlaps a measured silence, Vidiyo leaves that spot alone and tells the assistant. Gaps that aren't confirmed as silent stay for review. That's why the result sounds natural instead of chopped.

## How are filler words and repeated takes removed?

Fillers are removed by word. The assistant reads the transcript, picks the words to delete (an "um", a doubled "I, I", a restarted sentence) and sends a delete for those word IDs. Vidiyo turns that into exact cuts. The rule it follows is simple: cut a filler only when the meaning survives.

For longer recordings, the assistant can run a story analysis first. It looks for repeated takes and picks the best one, plus false starts and tangents, each tied to word positions. It asks before cutting a whole idea, because that's an editorial call, not cleanup.

## Can I control how tight the cuts are?

Yes, in plain words. "Only cut pauses longer than a second" keeps a relaxed rhythm, while "tighten every pause over a quarter of a second" gives the fast pace of a short. You can limit it to a section ("only in the first 20 seconds") or ask it to include the gaps it wasn't sure about. Because each request is one undo step, the quickest way to find your pace is to try one setting, watch the preview under the reply, undo and try another. Speech padding stays in every case, so even tight cuts don't clip the ends of words.

## What doesn't it do?

It doesn't speed up slow sections, because Vidiyo has no speed control. It doesn't add a minimum length or pad the video to hit a target: the duration follows what you said. And it can't cut a filler the transcript didn't catch by word. For those, press B to split at the playhead and delete the piece by hand.

Removing pauses creates [jump cuts](https://getvidiyo.app/glossary/jump-cut). Some creators like them raw. If you don't, ask for keyframed [punch-ins](https://getvidiyo.app/glossary/punch-in-zoom) on alternate cuts, or B-roll from the included stock library over the joins. Zooms are keyframed, not cursor-follow.

## Who is this for?

It's built for people who talk to camera: founders recording launch updates, creators making shorts, and anyone who records one long take and wants the pauses gone. It's also the first step of most edits in Vidiyo. Once the cut is tight, [auto captions](https://getvidiyo.app/features/auto-captions) and music follow in the same conversation.

## Honest limits

- A filler word can only be cut by word if it appears in the transcript. Anything missed can be split and deleted by hand.
- In a noisy room a pause may not measure as silent. Those gaps are left for you to review instead of being cut.
- Timelines max out at 5 minutes, and there's no speed control to speed up slow sections.

## Frequently asked questions

### Can Vidiyo remove silences from a video automatically?

Yes. Ask the assistant to "cut the long pauses" and it removes pauses that are measured as silent in the audio, keeping a little padding so speech doesn't sound clipped. The whole request undoes in one step.

### Does Vidiyo remove filler words like um and uh?

Yes. The assistant finds fillers in the word-timed transcript and deletes them by word, so cuts land between words rather than mid-syllable. It only cuts a filler when the sentence still makes sense without it.

### Can it remove repeated takes and false starts?

Yes. Ask "keep only the best take of each line" and the assistant reads the transcript for repeated takes, false starts and tangents, then cuts them. It asks before removing a whole idea.

### How do I hide the jump cuts after removing silences?

Ask for punch-ins. Vidiyo can push in slightly on alternate cuts so a jump cut reads as a new angle, or cover it with B-roll from stock footage. Its quality check flags raw jump cuts it finds.

### Does removing silences happen on my Mac?

Yes. Silence detection and transcription both run locally. The assistant receives the transcript and measurements to decide what to cut, through your Claude or ChatGPT plan, but your audio stays on your Mac.

## Related

- [Edit video by chatting](https://getvidiyo.app/features/chat-video-editing)
- [Auto captions that are transcribed on your Mac](https://getvidiyo.app/features/auto-captions)
- [Dead air](https://getvidiyo.app/glossary/dead-air)
- [Jump cut](https://getvidiyo.app/glossary/jump-cut)
- [Punch-in zoom](https://getvidiyo.app/glossary/punch-in-zoom)
- [How to remove silences from a video on Mac](https://getvidiyo.app/blog/remove-silences-from-video-mac)

Download Vidiyo for Mac (free during the beta): https://getvidiyo.app/download