How to cut the pauses and stutters out of a raw talking-head video

Last updated 2026-08-16 · 6 min read

What is the short version?

A raw take is mostly good with dead space between the good parts. Removing that by hand means scrubbing a two-minute recording for every breath and every restarted sentence, which is an hour of work for a video that is still two minutes long.

1. Go to Vizard Agent and upload the raw recording.
2. Say what to remove: silences, stutters, filler words, false starts.
3. Say whether you want captions burned on afterwards.

What do you need before you start?

The recording and a threshold. Vizard Agent measures the pauses itself, so what helps is knowing how aggressive you want the cut to be — a tight edit and a natural one differ by about a tenth of a second in what counts as a pause worth removing.

• The raw take. One continuous recording is the normal case.
• What counts as removable. Silence, "um", repeated words, sentences you started twice.
• How tight. Aggressive reads as energetic; too aggressive reads as breathless.
• Whether captions come after. They should be timed to the edited version, not the original.
• Anything to keep. A deliberate pause for effect is not dead air. Say if there is one.

What do you type into Vizard Agent?

Say what to remove and how tight to go. Vizard Agent finds the pauses by measurement rather than by ear, so the instruction that matters is where the line sits between a breath worth keeping and a gap worth cutting.

Prompt

Variants worth knowing:

• Both versions. Ask for the cut with and without captions; both are delivered as separate files.
• A gentler pass. "Only cut pauses over half a second" keeps the delivery natural.
• Remove a section too. Naming a passage to drop entirely fits in the same job.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real jump-cut edit. Every cut point in it comes from a number — a word timing, a gap length, or an audio energy measurement — rather than from a judgement about where the speech seems to stop.

1. Transcribes the video to word-level timings and inspects the transcript's structure.
2. Lists every segment with its timestamps to see the shape of the recording.
3. Examines the individual word timings segment by segment.
4. Compares first and last word timings in each segment to find the pauses between them.
5. Extracts the audio as a high-rate file for precise silence detection.
6. Measures audio energy around each pause to find where speech actually resumes.
7. Scans for consecutive repeated words — the stutters inside the parts being kept.
8. Finds every gap between words longer than 0.2 seconds.
9. Cuts frame-accurately and concatenates, then maps the word timings onto the new timeline for the captions.

Step 9 is the part that is easy to get wrong. After cutting, every word timing has moved; Vizard Agent recalculated them against the edited timeline so the captions match the new video rather than the old one.

What does the result look like?

From the run this page is written from, probed on the delivered file: 1080x1920, H.264, 30fps, 123.91 seconds, AAC audio. Vertical, just over two minutes after the dead air came out, with styled captions burned on and timed to the edited version.

Vizard Agent delivers both the cut without captions and the finished file, so you can take the clean edit into your own timeline if you would rather caption it yourself.

Expect tens of minutes rather than minutes. This category was not separately measured; comparable work runs a median of 28 to 38 minutes end to end, and across all projects the median cost by tier is Flash 47, Pro 55, Max 242, Ultra 263 credits.

When does this not work well?

Cutting silence changes the rhythm of someone's speech, and rhythm is a large part of how convincing they sound. Vizard Agent measures precisely and follows your threshold, and there is a point where a tighter cut makes a worse video.

• Too tight sounds breathless. Some pause is how people signal they have finished a thought.
• A jump cut is visible. The speaker's head jumps between cuts. That is the format, and it is why B-roll exists.
• Music or room tone breaks at the joins. Continuous background sound behind the speech will stutter where cuts land.
• Deliberate pauses get removed. Vizard Agent cannot tell a dramatic beat from hesitation. Say if there is one.
• Captions must follow the cut. Captions timed to the original will drift; they have to be built after the edit rather than reused from before it.
• One bad take stays a bad take. Removing the pauses tightens a recording; it cannot rescue a delivery that never worked.

How do you fix a result that came back wrong?

Say which way. Vizard Agent keeps the original transcript, the gap measurements, the audio analysis and the cut list, so loosening the threshold or restoring one pause is a recut from measurements it already has rather than a fresh analysis.

• "It feels rushed." A longer threshold, so shorter pauses survive.
• "Keep the pause before the last line." That one gap restored.
• "Bigger captions." A restyle over the same edit.

How does Vizard Agent compare to doing it yourself?

By hand this is scrubbing a waveform for every gap, cutting, listening back, and then finding you removed a breath that actually mattered. Editors do it well and it is reliably the least creative hour in the whole process, repeated for every take you record.

Common questions

Will it cut into my words? No. Vizard Agent measures the audio energy around each pause to find where speech actually resumes before it cuts.

Can it remove filler words specifically? Yes. Vizard Agent scans the transcript for repeated words and fillers separately from the silence detection, so you can ask for one without the other.

Do I get the version without captions? Yes. Both files are delivered, so you can take the clean cut into your own editor if you would rather caption or grade it there.

How much shorter will it be? It depends on the take. A nervous first recording loses far more to pauses and restarts than a rehearsed one does, and Vizard Agent reports the edited duration when it delivers.

Can it add B-roll over the cuts? Yes. Ask in the same conversation and Vizard Agent covers the jumps rather than leaving them visible.

Will the captions match after cutting? Yes. Vizard Agent recalculates the word timings against the edited timeline rather than reusing the original ones.

Can it keep the video wide? Yes. Vizard Agent delivers in whatever shape you ask for and defaults to the source's.

Does it re-transcribe after cutting? No need. Vizard Agent maps the original word timings onto the new timeline instead of transcribing twice.