How to Remove Filler Words from Video Without Awkward Cuts

Published by Rescreen, which makes one of the products discussed. How we source and update these guides.

To remove filler words from a video, identify the unwanted “um” or “uh,” preview the surrounding sentence, and cut the corresponding media in your editor. Replay the join with the picture visible before applying the same operation elsewhere. A transcript can help you find candidates, but deleting a word from captions alone will not necessarily remove it from the recording. To get a transcript without uploading the recording, the free video to text converter makes one in your browser, with a time on every line.

For a screen walkthrough, the best edit keeps both the explanation and the action it describes. A hesitation may happen while a menu opens or a result appears. Removing that interval without checking the picture can leave a smoother voice explaining a step the viewer never sees. This guide shows how to review candidates, use the documented Descript and Adobe Premiere controls, and repair a cut that does not work.

The product instructions below were checked against linked vendor documentation on 15 October 2026. The practice script is an original fictional exercise, not a customer result or hands-on product benchmark. You do not need to remove every hesitation to finish a useful video. You need a clear explanation that still sounds like the person who recorded it.

Choose which words to remove

Begin by listening to the whole explanation once. Mark moments that interrupt understanding, rather than searching for a maximum number of deletions. An isolated hesitation between complete phrases is a different editing problem from a sound attached to the beginning of a word. Treat an automated list as a set of candidates, not an instruction to approve every cut.

For example, removing “like” from “choose a format like MP4” changes the sentence. Similarly, a speaker saying “I think” may be expressing uncertainty rather than adding empty sound. Do not use cleanup to turn a qualified statement into a confident one. If the speaker corrects a number or instruction, identify the intended final statement before editing the false start.

Decide what the viewer must learn from each section. In a short software lesson, that might be where to open a setting, what value to enter and how to recognize success. Keep that list beside the edit. It is a more useful acceptance test than reaching zero highlighted fillers.

Prepare a reversible edit

Keep an untouched copy of the source recording and save a separate working project or sequence. Give the first export a different filename from the original. If you already have captions, retain their original version too: cuts can change timing, and a later revision may need material you removed from the first pass.

Check the transcript against the recording around names, numbers and commands. A wrong transcript word is a text-correction problem; unwanted recorded speech is a media-editing problem. Decide which one you are solving before choosing an operation. Listen to one short section and confirm that the editor's playback position matches the words and picture you expect.

Then make one trial cut in a low-risk passage. Save, play it in context and undo it once so you know how to recover. Avoid testing a bulk action on the only working version of a long recording. The small trial establishes what the selected control changes before you apply it to important material.

If you are still choosing software, use the Descript alternatives guide for workflow differences. This article is an editing procedure for an existing recording, not a ranking of paid products or a recommendation to buy another subscription.

Remove filler words in Descript

Descript's filler-word documentation currently specifies English transcripts and AI-credit usage on current plans. Check access in your account before relying on a bulk cleanup.

1. Open the composition containing your transcribed recording.
2. Open AI Tools in the right sidebar and choose Remove filler words.
3. Preview a detected instance and choose the intended operation. Ignore leaves a struck-through record while removing the audio; Delete removes text and audio. Remove from transcript leaves the audio unchanged.
4. Enable Avoid harsh cuts when appropriate. The documented behavior is to skip candidates that would clip nearby words or produce awkward joins; still listen yourself.
5. Replay the sentence with the picture visible. Restore a bad edit before continuing.

The same documentation describes Delete and replace with gap for retaining timing with a gap clip. Preserving duration is not proof that the necessary picture remains intact. Inspect the result. For a restricted section, Descript documents duplicating the script selection into a new composition, processing it there, then copying the cleaned script back.

Keep pause shortening as a separate pass. Descript's Shorten word gaps instructions let you define qualifying gaps and a target length, then shorten one or all matches. Start with individual results. A pause that feels unnecessary in audio may still give the viewer time to locate a small control. There is no single target duration that proves every tutorial is easy to follow. To cut long silences on their own, without a transcript, the free video silence remover shows every pause before it cuts and leaves filler words alone.

Find filler words in Premiere

Adobe's current Text-Based Editing instructions describe filtering transcript results for filler words and deleting individual or multiple matches.

1. Import the recording into a project and open the Text-Based Editing workspace.
2. With the transcript available, select the filter icon in the Transcript panel.
3. Choose Filler words. Keep pause removal as a separate decision.
4. Preview a result and use Delete for a single instance before considering Delete all.
5. Review the edited sequence, including picture, sound and any other material that crosses the cut.

Use the sequence as the final check, not the disappearance of a search result. After a trial deletion, compare the before-and-after timing and make sure the expected media changed. Save the working version before extending the operation across a longer section.

If a tutorial or installed version presents different controls, do not select a vaguely similar option and assume it performs the same edit. Consult the linked current documentation, identify whether you are viewing source text or an edited sequence, and test one reversible change. The expected outcome is a reviewed media edit, not merely a cleaner-looking transcript.

Make a careful manual cut

You can use the same review process in a timeline editor without automatic filler detection. Find the unwanted sound by listening, zoom in around it and identify its start and end. The waveform can help locate a boundary, but it does not tell you whether a sound carries meaning or whether the picture contains an important action.

1. Play from the beginning of the phrase and mark the candidate sound.
2. Make a reversible selection around that sound, leaving the neighboring words intact.
3. Apply the editor's intended media cut, checking which audio and video tracks are affected.
4. Play across the join at normal speed, beginning before the preceding word.
5. If the sentence is clipped or the picture skips a required step, undo and choose a different boundary or leave the hesitation.

Do not remove a span from only one synchronized track by accident. If you intentionally edit narration separately from the picture, check alignment at a later visible action as well as immediately after the cut. A locally smooth join can still leave a later instruction describing the wrong frame.

When a filler is fused with a useful word, keeping it can be the better edit. Another option is a pickup of the whole phrase or screen step. Record a complete replacement that can be reviewed in context, rather than trying to stitch several tiny fragments into speech that was never naturally spoken.

Practice without losing a screen action

Try this fictional practice script on a disposable demonstration: “Open Export. Um, choose the smaller size. Wait for the preview. Now save the file.” Perform each action as you speak. During the hesitation, deliberately leave the size menu open so there is something meaningful on screen to inspect.

First make a speech-focused edit that removes only the hesitation. Watch it without looking at the transcript. Can you still see where the size option is and which item was chosen? Then play it with sound. Does the instruction arrive before or while the relevant action is visible? A shortened recording passes only when both checks succeed.

Now consider a harder variation: the menu opens and closes entirely during the hesitation. A cut of the whole interval may hide the only evidence of the selection. Keeping that interval, changing the edit boundary or recording the step again are honest repairs. A decorative transition cannot restore an action that is no longer present.

Keep the untouched take and your edited version side by side. This is a proposed exercise, not a claim about how either named editor performs. Use your own results to decide which automatic changes are safe to keep.

Repair awkward cuts

These checks identify where to investigate; they do not prove that every matching symptom has the same cause. Keep one failed sample so you can compare it with the next attempt. Changing transcription settings, cut boundaries and export choices together makes it difficult to know which change repaired the problem.

If a sentence sounds chopped, undo the last change and replay the original boundary. Listen for a missing consonant, an unfinished breath or a word that was mistaken for filler. Change one boundary at a time and compare again. If you cannot make a clean join, retain the original phrase or record a replacement.

If the picture jumps, ask whether it is merely noticeable or whether it hides information. A speaker changing posture is different from a missing menu selection. For a tutorial, prioritize the viewer's ability to reproduce the step. Recheck captions after the repair; the text and its timing should describe the final edit. If your player needs a different caption format from the one your editor exports, the free SRT to VTT converter converts it without changing a timing.

Check the exported video

Save the working project and export the intended edited version. Open the resulting file outside the editor, with headphones, and watch from start to finish at normal speed. Do not rely only on the editor preview or the transcript. Verify that the delivered file contains the changes you approved.

• Meaning: qualifications, numbers and instructions still express the original intended message.
• Sound: neighboring words are complete and the joins are comfortable to listen to.
• Picture: each required action and result remain visible.
• Timing: narration, captions and picture agree throughout the file.
• Delivery: you have the right version and the recipient can open it.

Record the decisions you may need to reverse

Download the filler-word edit review log (CSV). Enter the source timecode, words around the cut, intended operation and reason for the edit. The observation fields are blank for your own checks; this is a worksheet, not published test data.

Before exporting a long finished lesson, make a short review export containing a representative cut and the next required screen action. Compare it with the working sequence. If it passes, export and review the complete lesson as described above. The sample catches a wrong version or operation early; it does not replace watching the full delivery file.

In the log, describe a failure precisely: “the first sound of choose is missing” or “the size menu closes before it can be read.” Record whether you restored the original, changed a boundary or recorded a pickup. This makes feedback actionable when another reviewer finds the same moment, and prevents a later cleanup pass from repeating an edit you already rejected.

Keep the source and project until the finished video has been reviewed. If the lesson still feels rushed, revisit the pauses around new information before removing more speech. For planning the larger lesson, use the tutorial-video workflow; for a closer look at Descript's overall fit, see the Descript evaluation guide.