How to remove filler words from an interview video

To remove filler words from a video interview, transcribe the footage, search the transcript for "um", "uh", "like", "you know" and false starts, mark them as a single candidate edit, then play every join before you accept it. The transcript finds the words quickly; your ears decide which ones go. Most of the quality comes from what you keep: a pause before an honest answer, a stumble that shows someone thinking, a breath that gives the audience time to catch up.

This guide covers the method in any editor that can transcribe, the habits that stop a cleaned interview from sounding robotic, and how to do it locally in cinemaOS with J.A.R.E.D.

Why filler words are worth removing, and why not all of them

Spoken language is full of hesitation. In a room, nobody notices. On screen, three "ums" in a ten-second answer make a confident subject look unsure, and they make a three-minute piece feel like four.

But filler is not noise to be deleted on sight. It often carries information:

• Thinking time. A pause before a hard question tells the viewer the answer is considered, not rehearsed.
• Emotion. A broken start ("I, I didn't...") can be the most truthful moment in the interview.
• Rhythm. Natural speech has breathing room. Strip all of it out and the subject sounds like they are reading from a teleprompter at speed.
• Meaning. "Like" is sometimes a filler and sometimes the verb. "So" can be a habit or the start of a conclusion.

The aim is a tighter, clearer version of what the person actually said, not a different person.

The method, in any editor

The basic workflow is the same whatever software you use. Tools differ in how much of it they automate.

1. Transcribe the interview. Use a transcript that is timed to the media, so selecting a word selects the matching moment in the clip.
2. Search for the usual suspects. Look for "um", "uh", "er", "erm", "you know", "I mean", "sort of", "kind of" and repeated words such as "the the". Note that some transcription models tidy filler out of the text by default, so a clean-looking transcript does not mean clean audio.
3. Find false starts by reading. A false start is a sentence the speaker abandons and restarts: "We went to, we drove to the coast." Search will not find these. Reading will.
4. Mark, do not delete yet. Gather your removals as a set you can review together, on a duplicate sequence or as a proposed edit, so the original stays intact.
5. Listen to every join. Play a second or two either side of each cut. You are listening for clipped consonants, breaths cut in half, a jump in room tone and a change in pitch that makes the cut obvious.
6. Cover what you cannot hide. If a cut is clean in sound but jumps in picture, use B-roll, a cutaway to the interviewer, or an honest jump cut.

Several editors now help with steps 2 and 4. Descript has a dedicated filler word removal feature built around its transcript. In recent versions, Premiere Pro's text-based editing can flag filler words such as "um" and "uh" and pauses in the transcript so you can delete them. DaVinci Resolve has added transcription and text-based editing tools in recent versions, with several of them limited to the paid Studio edition. Features change between releases, so check each vendor's current documentation for what your version does. The review in step 5 is the same in all of them.

What to keep: pauses that carry meaning

The hardest part of filler removal is restraint. A few rules of thumb hold up well on real interviews:

• Keep the pause before a difficult answer. If the question is personal, the silence is part of the answer.
• Keep a hesitation when the subject is correcting themselves on purpose. "It was, well, it was more complicated than that" is not a false start. It is the point.
• Cut the filler, not the breath. Removing an "um" and the breath that follows it in one cut often leaves the next word starting unnaturally fast. Leave a short breath in.
• Leave a little room after a laugh. Laughter decays slowly. Cutting straight to the next sentence clips the tail and sounds wrong.
• Watch the face as well. A cut that is clean in audio can still show a head jump or a blink that gives it away.

As a rough approach, do one pass that removes obvious filler and false starts, then a second pass where you put back anything that now sounds rushed. The second pass is where an edit starts to sound human.

Removing filler words in cinemaOS with J.A.R.E.D.

cinemaOS is a Mac-native editor for Apple silicon. Transcription and its assistant, J.A.R.E.D., run on your Mac, so the interview, the transcript and your requests are not uploaded, and there are no generative credits to count. The steps below follow the transcribe and edit by text tutorial.

Step 1: Transcribe the interview locally

Select the interview in the Sequencer and press Shift Command T to transcribe with J.A.R.E.D. The transcript is timestamped and can assign up to four speakers, which covers most interview setups. Correct names and speaker labels before you go further, because the same transcript feeds your captions later.

Step 2: Read it in the Text Editing tab

Open the Text Editing tab in the bin and read the words alongside the footage. Selecting a passage and removing it cuts the linked picture and sound clips on the timeline. This is where you find false starts by eye: read each answer and look for sentences the speaker abandons.

Step 3: Ask J.A.R.E.D. for a candidate, not a final cut

Press Command J and ask for something like "remove filler words as a candidate". J.A.R.E.D. treats subjective editorial work as an editable candidate: it does not silently replace a sequence you have already approved. You get a filler-free candidate to review while the approved sequence stays as it was, and when changes land on the timeline, a receipt of what changed. Check both what the candidate removed and what it left in. False starts and judgement calls still come from your own reading in Step 2.

Step 4: Review, restore and undo

Play the candidate through at normal speed. Where a removal sounds wrong, put it back, or undo the step with Command Z, which uses the app's ordinary Undo and Redo. Projects are local .mntm files with version snapshots and recovery, so you can also keep a snapshot of the uncut interview before you start and compare against it.

Step 5: Finish the sound and caption

Small cuts in dialogue can leave clicks and room-tone jumps, so fix those first with the techniques in the next section. Then balance the dialogue bus in the SOUND console and check loudness before export. When the cut is locked, generate captions from the same transcript; the offline subtitles tutorial shows how.

What cinemaOS does not do here

Be realistic about the limits. J.A.R.E.D. runs only the jobs it has been built for and refuses requests outside them, rather than guessing. Transcript accuracy and speaker detection depend on the recording, so noisy or overlapping speech needs more manual review. Treat the filler candidate as a first pass, not a complete one: phrases like "you know" or "I mean", and filler the transcript did not write down, may still need a manual pass in the Text Editing tab. The candidate also does not judge which hesitation is meaningful. That decision stays with you. cinemaOS is Mac-only for now, with Windows planned for November 2027, and it is a young app at version 0.2, Build 9.

Common problems after removing filler

• Clicks at the cut. Add a short audio crossfade on each join, or nudge the cut point to a zero crossing between words.
• Room tone jumps. Fill gaps with a few frames of clean room tone from the same take rather than leaving digital silence.
• The subject sounds rushed. You removed breaths along with the filler. Restore a few, starting with the ones before key sentences.
• A false start was the better line. People often say the clearest version first and then over-explain. Read both before choosing.
• Captions no longer match. Regenerate or re-time captions after the cut is locked, not before.

If you are reviewing an automated edit for the first time, when the AI editor says done but the cut is wrong covers how to check machine edits against your own judgement.

FAQ

Can I remove filler words automatically without reviewing them?

You can in several tools, but it is rarely a good idea for interviews. Automatic removal treats every "um" the same, and some of them carry meaning or protect the rhythm of speech. Treat any automatic pass as a proposal and listen to the result.

Does removing filler words change what the speaker meant?

It can. Cutting a hesitation can make a tentative answer sound certain, and joining two halves of a sentence can create a statement nobody made. If an edit changes the meaning, put it back, especially in documentary and news work.

Is it better to remove filler words before or after the story edit?

After. Choose your answers and build the structure first, then clean filler only in the lines that survive. Cleaning a two-hour transcript that will become three minutes wastes time. For the story edit itself, transcribe an interview and edit by text walks through the full workflow.

Try it on one answer

Pick a single answer from your next interview, clean it by hand, then compare it with an automated candidate. You will quickly learn which hesitations you want to keep. If you edit on an Apple silicon Mac and want the transcript and the assistant to stay on your machine, cinemaOS is free for 60 days with no card and no automatic charge, then $9.99 USD a month excluding VAT.