How to Remove Um and Uh From a Recording

Um and uh sneak into every kind of recording — voice memos, client calls, voiceovers, Loom walkthroughs, classroom lectures. You sound fine in your head while talking, then play it back and hear yourself stalling forty times. The good news: you do not need to re-record. Modern filler word remover tools transcribe your audio, highlight every um and uh automatically, and let you delete them the way you would delete words in a document.

The three ways to get rid of um and uh

1. Scrub the waveform by hand. Zoom into the timeline, find each filler, cut it. This works in any editor and costs nothing, but a ten-minute recording can take an hour of hunting. Viable only for short clips.

2. Use an AI filler word remover. The tool transcribes the recording with a timestamp on every word, flags the fillers, and removes them in bulk. What took an hour now takes minutes. This is the approach UmBreak takes: it runs entirely in your browser, so your recording never gets uploaded anywhere.

3. Record it again, better. Always an option, and sometimes the right one — but re-recording costs time you may not have, and the new take will have its own ums. A hybrid works best: clean what you have with AI, then work on saying fewer fillers next time.

Removing um and uh with UmBreak (step by step)

1. Get your audio into one file. Export from your recorder or editor as MP3 or WAV. For video, export the video file — UmBreak transcribes the audio track from MP4s too. Up to about 30 minutes works in one pass.
2. Drop it into the tool. Open the filler word remover and drop the file in. Your browser decodes it locally — nothing leaves your device, which matters if the recording is a client call or an unreleased project.
3. Transcribe on-device. Click transcribe. An AI speech model (downloaded once, about 40 MB, cached after) timestamps every word. Skim the transcript for mistakes, especially names, jargon, and brand terms.
4. Remove the ums and uhs. They are highlighted automatically. Click Remove all filler words for the bulk pass, then click individual words back in if any of them sound natural — an occasional filler keeps speech human.
5. Preview, then export. Play the cleaned version with live word highlighting. When it sounds right, download the WAV and drop it into your editor or publish it directly.

Most short recordings are done in under five minutes.

Which ums and uhs to keep

Zero-fillers speech sounds robotic. Keep the fillers that do conversational work: a thoughtful “uh” before a considered answer, a filler that buys a beat in storytelling. Cut the clusters (three ums in one sentence), the ones at the start of every answer, and the ones that repeat in the same cadence — those are the ones listeners notice. Listen to the preview before exporting; your ear is the final judge, and you can restore any word with a click.

Honest limits of this method

Two things AI filler removal cannot do well. First, transcription is not perfect: if the model mishears a real word as a filler, review the highlights before bulk-removing, especially with accented speech, technical vocabulary, or background noise. Second, fillers buried in crosstalk — two people talking over each other — are hard for any tool to cut cleanly without a rough edit point. For clean solo recordings, though, the result is nearly seamless.

If the recording has other problems too

Um and uh rarely travel alone. Recordings with fillers usually also have long dead-air gaps and background hum. Handle them in this order: remove fillers first, then trim the worst silences (UmBreak has a dedicated awkward-pause removal guide), and if the background is noisy, run the noise remover before transcribing — cleaner audio transcribes more accurately, which makes the filler detection better too.