How to Remove Ums and Uhs From Audio for Free

Ums and uhs are the most common junk in spoken audio. Podcasts, voiceovers, lectures, client calls — they all collect these little stalls, and they are the first thing listeners notice on playback. The good news is that cleaning them out costs nothing. The question is which free method fits your recording, because "free" covers everything from an hour of manual cutting to one-click AI.

What your free options actually are

1. Manual cutting in a free editor. Audacity, CapCut, DaVinci Resolve: zoom into the waveform, find each um, cut it. Costs nothing and teaches you a lot about editing. But a ten-minute episode with forty ums becomes an hour of hunting and splicing, and the cuts sound choppy unless you add tiny crossfades. Fine for a two-minute clip; miserable for a weekly show.

2. Free tiers of AI tools. Several paid apps that remove um and uh automatically offer free plans with time or feature caps. They work well, but the caps are designed to convert you: limited minutes per month, watermarks on exports, or your audio uploaded to their servers for processing.

3. In-browser tools with no account. This is the approach UmBreak takes. The AI transcription model downloads into your browser (about 40 MB, cached after the first run) and timestamps every word on your own device. Ums and uhs get highlighted automatically, you delete them like text, and you export the cleaned audio. Nothing is uploaded anywhere, so there is no server to pay for and no account to make.

The free no-signup workflow (step by step)

1. Export one clean audio file. From your recorder or editor, export as MP3 or WAV. For video projects, export the MP4 — the audio track is transcribed from it. Recordings up to about 30 minutes work in a single pass.
2. Open the tool and drop the file in. The filler word remover runs entirely in your browser. Your audio is decoded locally; it never leaves your device, which is worth something when the recording is a client call or an unreleased episode.
3. Transcribe on-device. Click transcribe and skim the result. AI transcription is good, not perfect: check names, jargon, and brand terms before you trust the highlights. If the audio is noisy, run the noise remover first — cleaner audio transcribes more accurately, which makes filler detection better.
4. Remove the ums and uhs. They are already highlighted. Use the bulk removal for the obvious clusters, then review individually: an occasional filler between sentences sounds human, while three ums in one answer sound like a glitch. Restore any word with a click.
5. Preview and export. Play the cleaned version with live word highlighting. Listen for rough edit points, especially where a filler sat inside a breath. When it sounds right, download the WAV and drop it back into your editor.

Which ums and uhs to keep

Stripping every filler makes speech sound robotic. Keep the ones doing conversational work: a thoughtful "uh" before a considered answer, a filler that holds a beat in a story. Cut the clusters, the ums that open every answer, and the ones repeating in the same cadence — those are the ones listeners count. The preview is your final judge, and dead-air gaps nearby often matter as much as the fillers themselves.

Honest limits of the free approach

Three things to know before you start. First, transcription errors happen: accented speech, technical vocabulary, and crosstalk get misheard, so review the highlights before bulk-removing. Second, two people talking over each other is hard for any tool to cut cleanly; solo recordings get nearly seamless results, messy panel discussions do not. Third, free in-browser tools run on your hardware, so a very long or very noisy file may take longer to process than a server-side service would. For a typical podcast episode, though, you will be done in minutes instead of the hour manual editing costs.