UmBreak
Free · No signup · 100% in-browser

Remove the ums. Keep the voice.

UmBreak is a free filler word remover that transcribes your audio on your own device, highlights every filler word — um, uh, like, you know — and cuts them out with one click. Your file is never uploaded anywhere. No account, no watermark, no catch.

umbreak — local session Private by design

Drop your audio or video file here

or click to browse your device. The file never leaves this tab.

MP3 · WAV · M4A · OGG · MP4 · WEBM · up to 30 min

Reading your file

0%

First run downloads a ~40 MB speech model from a public CDN (one time, cached afterwards). Your audio is never sent anywhere — transcription runs on your device.

Detected filler — click to keep Cut — click to restore Silence removed
0
Words transcribed
0
Fillers found
0
Words cut
0:00 saved
Audio trimmed

Transcription is done by an on-device AI model (Whisper, tiny). It is good but not perfect — skim the highlighted words before exporting.

One tab. Three free tools.

UmBreak is a free creator toolkit. Every tool below runs 100% in your browser — no upload, no signup, no watermark.

Audio

Filler-word remover

Transcribes your audio on-device, highlights um, uh, like and you know, and cuts them with one click. Now with silence removal and SRT export.

Open the tool → Audio & video

Noise remover

Strips background hiss, hum, and fan noise from recordings. Compare original vs cleaned, tune the strength, download a clean WAV.

Open the tool → Video

Video cropper

Crop any video to 16:9, 9:16, 1:1, or a custom frame — right in your browser. Drag the crop area, preview, and export.

Open the tool →

How it works

Three steps. Everything happens inside your browser tab — there is no server processing your voice, because there is no server involved at all.

Step 1

Drop your file

MP3, WAV, M4A, or even MP4 video — the audio track is read locally. Nothing is uploaded, ever.

Step 2

AI transcribes on-device

A speech-recognition model runs in your browser and timestamps every word. First visit downloads the model once (~40 MB), then it works offline.

Step 3

Cut, preview, download

Filler words are highlighted automatically. Click any word to cut or keep it, preview the result, and download a clean WAV.

Private by design, not by promise

Most audio tools make you upload your voice to their servers and create an account first. UmBreak is built the other way around: your file never leaves your device, so there is nothing to leak, sell, or subpoena.

What never happens

  • Your audio is never uploaded. Transcription runs in this tab via WebAssembly.
  • No account, no email, no tracking pixels tied to your files.
  • Closing the tab wipes everything. We keep no copies because we never receive any.
Abstract dark studio waveform artwork representing on-device audio editing
UmBreak processes audio where it belongs: on your device.

Read the full privacy policy →

Filler words we catch

Two groups. Classic fillers are almost always safe to cut. Discourse markers depend on context — UmBreak flags them, you make the call. Toggle either group in the tool above.

GroupExamplesDefaultNotes
Classic fillersum umm uh uhh er ah hmmOnHesitation sounds. Cutting them is nearly always an improvement.
Discourse markerslike you know i mean basically actually literally sort of kind ofOnContext-dependent. “Like” can be a verb — review before removing.

Questions, answered honestly

The things people actually ask before trusting a tool with their voice.

How do I remove filler words from an audio file?+
Drop the file into the tool above. UmBreak transcribes it on your device, highlights every um, uh, like, you know and similar filler, and lets you cut them with one click — or click individual words for precise control. Preview the result, then download a cleaned WAV.
Is my audio uploaded anywhere?+
No. Transcription runs entirely in your browser using a downloaded AI model. Your file never leaves your device, there is no account, and closing the tab wipes everything. You can verify this yourself: disconnect from the internet after the model loads and the tool keeps working.
Which languages are supported?+
English only, for now. The built-in model is an English-only speech model. Multilingual transcription is on the roadmap, but we would rather do one language well than five badly.
How accurate is the transcription?+
It uses Whisper’s tiny model, which is fast and small (~40 MB) but not the most accurate variant. Clear speech in a quiet room transcribes well; heavy accents, crosstalk, or background noise will produce errors. Always skim the highlighted words before exporting — that is what the preview is for.
Does it work with video files?+
Partially. You can drop in an MP4 or WebM and UmBreak will transcribe its audio track and export a cleaned audio file. Re-attaching the cleaned audio to your video (re-muxing) is not supported yet — that is planned for a later version, and this page will say so when it lands.
Is it really free? What is the catch?+
Free, no catch. There is no server to pay for because all the work happens on your device — that is also why no signup is needed. If a paid tier ever appears, it will be for extras (like batch processing), never for the core filler remover.
What are the limits?+
Files up to 30 minutes. English only. Transcription quality depends on your audio — it is AI, not magic. And very long files take a while on slower machines, since your CPU is doing the work.
Which browsers work?+
Recent Chrome, Edge, Firefox, and Safari on desktop. The tool needs WebAssembly and the Web Audio API; most browsers from the last few years qualify. Mobile browsers can work for short files but are not the focus yet.
How do I remove filler words from a podcast?+
Export your episode as an audio file (MP3 or WAV) and drop it into the tool above. UmBreak transcribes it on your device, highlights every um, uh, like, you know and similar filler, and lets you remove them all at once or keep the ones that sound natural. Preview the cleaned version, download the WAV, and drop it back into your podcast editor. There is a full step-by-step guide here. One honest note: transcription is AI-based, so skim the highlighted words before exporting — especially names and jargon.
How do I remove um and uh from a recording?+
Um and uh belong to the tool’s “classic fillers” group, which is switched on by default. Upload your recording, run the transcription, then click Remove all filler words. Every um and uh is cut from the audio with a small safety margin so no speech gets clipped. Listen to the preview first — if a hesitation actually sounds natural in context, click that word to keep it.
Is there an app that removes um and uh automatically?+
UmBreak does exactly this, but as a website rather than an installed app — open it in your browser, drop in a file, and the AI finds and cuts um and uh automatically. Nothing to install, no account to create, and your audio never leaves your device. It works on desktop browsers; a native mobile app is not planned.
Is UmBreak a free Descript alternative?+
For filler-word removal specifically, yes. UmBreak gives you the same core workflow — transcribe, highlight fillers, delete them from the text, export clean audio — free, with no account. Be aware of the difference: Descript is a full editing suite with multitrack timelines, publishing, and collaboration. UmBreak is a single-purpose free tool that does one job well.
Can I edit audio by editing text?+
Yes — the transcript is the editor. After transcription, every word in the transcript maps to a moment in your audio. Click any word to cut it from the audio, or click a cut word to restore it. This is the same text-based editing approach pro tools use, running entirely in your browser. One caveat: word timestamps from the on-device model are approximate, so cuts include a small safety margin.
Is there a free podcast audio cleaner with no signup?+
That is what UmBreak is. No signup, no email, no account — open the page and start cleaning. It is free because there is no server processing your audio; your own device does the work. Current limits: files up to 30 minutes, English only, and transcription quality depends on how clean your recording is.
Can UmBreak remove long silences and dead air?+
Yes. After transcription, click Remove silences and the tool cuts every pause longer than your chosen threshold (0.5 to 2 seconds), including dead air at the start and end of the recording. A short breath of air is kept at each cut so it does not sound robotic, and removed silences are marked in the transcript and on the waveform — preview before exporting, as always.
Does UmBreak normalize loudness?+
Optionally, at export. Switch on Normalize volume and the cleaned WAV is scaled so its loudest peak hits −1 dBFS — a simple level adjustment done on your device. One honest note: this is not AI audio enhancement. It will not remove background noise or make a bad mic sound like a studio — it just evens out the level.

Honest limits — read before you rely on it

  • AI transcription makes mistakes. Names, jargon, and accented speech are the usual casualties. Check important words.
  • Word timestamps are approximate. Cuts include a small safety margin so speech is never clipped, which means a sliver of silence can remain.
  • “Like” is context-dependent. The tool flags it; only you know whether it was a verb. Discourse markers are a suggestion, not a verdict.
  • Video export is audio-only for now. You get a cleaned WAV, not a re-cut video file.