Loading tool…
Ums, uhs and dead air make a good recording feel unpolished. This tool transcribes your audio or video with word timings, highlights the fillers and long silences it finds, and lets you choose exactly which ones to cut before it stitches everything back together.
Processing locally on your device
Overview
Filler removal starts with knowing where each word begins and ends. The tool runs OpenAI's Whisper (Apache-2.0) in your browser with word-level timestamps, then scans the words for hesitation sounds such as um, uh, er, ah and hmm (plus euh, äh and ähm), for immediately repeated words like “the the”, and optionally for filler phrases like “like”, “you know” and “I mean”. Phrases are shown as suggestions only, because they are often meaningful words.
Whisper has a habit worth knowing about: it was trained to write clean transcripts, so it often leaves fillers out. The tool deals with that in two ways. A verbatim hint primes the model with a disfluent English example so it writes down more hesitations. And it measures the audio level between recognised words: a gap that contains sound but no transcribed words is often a filler or false start, and is offered as an “untranscribed sound” you can choose to cut. Silences longer than a threshold you set are shortened to a natural length rather than removed outright.
Each cut is shown in the transcript and on the waveform. Click any highlighted word or pause to keep or remove it, or toggle a whole category at once. When you export, the kept parts are joined with short equal-power crossfades (5–60 ms) so the edits don't click. Audio is processed directly in the browser and exported as MP3, M4A, WAV, FLAC, OGG or Opus. For video, cut points are snapped to frame boundaries, the audio is spliced the same way, and FFmpeg compiled to WebAssembly re-encodes the picture to MP4 so sound and image stay in sync.
Without WebGPU, Whisper Base (77 MB) or Tiny (41 MB) runs on your CPU; with WebGPU, Small is available too. Expect transcription to take about as long as the recording on a typical laptop CPU, and much less on a GPU. Treat the result as a first pass and listen through: automatic timings are good but not perfect, and some pauses are there for emphasis.
| Type | Examples | Default |
|---|---|---|
| Filler sounds | um, uh, er, ah, hmm | Cut |
| Repeated words | the the, I think I think | Cut |
| Long pauses | Silences over 1 s, shortened to 0.4 s | Cut |
| Untranscribed sounds | Audible gaps between words | Suggested |
| Filler phrases | like, you know, I mean | Suggested |
Drop in an audio or video file.
Choose the Whisper model and language; keep the verbatim hint on for English.
Click Find filler words and wait for the transcription.
Review the highlighted words and pauses, clicking any you want to keep or cut.
Set the pause length and crossfade, then click Remove to export the edited file.
Use cases
Tighten an interview or solo episode in minutes instead of hunting for every um by eye in an editor's waveform.
Remove hesitations and long silences from recorded talks so they feel more confident and take less of the viewer's time.
Clean up a narration take without re-recording it: cut the stumbles and repeated words, keep the natural rhythm.
Trim filler and dead air from webcam or phone videos, then export an MP4 with cuts that stay in sync.
Pro tips
Answers to common questions about the filler word remover.
Whisper tends to tidy speech as it transcribes and often skips fillers. The verbatim hint helps, and the untranscribed-sound detector catches many of the rest, but expect to review the result by ear.
Each join gets a short equal-power crossfade, and pauses are shortened rather than removed, which keeps the rhythm natural. You can adjust the crossfade length.
Yes. The cuts are snapped to whole frames, the audio is spliced at the same points, and the video is re-encoded to MP4 so picture and sound stay in sync.
Detection of hesitations, repeats and pauses works for any language Whisper supports, and a few non-English hesitations are recognised. The verbatim hint and the filler-phrase list are English only.
Dashed items are suggestions that start switched off, such as “like” or untranscribed sounds. Click one to cut it.
No. ToolEnhancer processes files on your own device inside the browser. They are never sent to, stored on or seen by our servers.
Yes. The filler word remover is free to use with no sign-up, no watermarks and no daily limits.