Silence & Filler Word Remover

Cut dead air and “um”s from podcasts & voiceovers — silences detected, previewed and cleaned entirely on your device, never uploaded.

Cut dead air and “um”s from podcasts & voiceovers in your browser. Tunable silence detection, crossfaded joins, MP3/WAV + cut-list CSV. Free, no upload.

About Silence & Filler Word Remover

The Silence & Filler Word Remover cleans up podcasts, voiceovers and lecture recordings entirely in your browser. Drop in an audio or video file and it scans the loudness (RMS over ~50 ms windows) to find dead air, highlights every gap on a waveform, and lets you tune the threshold, minimum pause length and padding with live before/after preview. An optional on-device Whisper pass (desktop only) also flags standalone “um”, “uh”, “erm” and “hmm” as cuts. Kept segments are spliced with ~10 ms crossfades and exported as MP3 or WAV, plus a cut-list CSV — nothing is ever uploaded.

How to use Silence & Filler Word Remover

  1. Drop in an audio or video file (MP3, WAV, M4A, FLAC, MP4, MOV…) — it is decoded and analyzed on your device.
  2. Tune the three sliders: silence threshold (dB), minimum silence length, and the padding kept around speech so cuts never clip words.
  3. Review the highlighted gaps on the waveform and in the cut list — click any cut chip to keep that section instead.
  4. Optionally (on desktop) run “Also find filler words” — on-device Whisper transcribes the audio and adds standalone “um”/“uh”/“erm”/“hmm” as labeled cuts.
  5. A/B check your edit with Play original vs Play cleaned.
  6. Export the cleaned MP3 or WAV — and the cut-list CSV (start, end, type in seconds) if you want to apply the same cuts in a video editor.

Frequently asked questions

Is my audio uploaded anywhere?
No. The file is decoded, analyzed, previewed and re-encoded entirely in your browser with the Web Audio and WebCodecs APIs — nothing ever leaves your device, and there is no signup or watermark.
How does the silence detection work?
It measures loudness (RMS) over ~50 ms windows and marks any stretch quieter than your threshold (default −40 dB) that lasts longer than the minimum silence length (default 0.6 s) as a cut. A padding setting (default 0.15 s) keeps a little silence on each side of speech so words are never clipped, and dead air at the very start or end of the file is removed completely.
Can it jump-cut the silences out of my video?
Not yet — in this version a video file gives you the cleaned audio track (MP3/WAV) plus a cut-list CSV with every removed range in seconds, which you can apply as jump cuts in your video editor. The video itself is not re-rendered in the browser.
How are filler words like “um” and “uh” found?
An optional pass transcribes the audio with an open Whisper model running on your device (desktop browsers only — the model is too heavy for phones) and flags cues that are nothing but a standalone “um”, “uh”, “erm” or “hmm” as labeled, toggleable cuts. Your audio never leaves the page; only the model weights are downloaded once.
Will I hear clicks or pops where the audio was cut?
No — the kept segments are spliced with ~10 ms crossfades at every join, so the waveform never jumps abruptly. Use the padding slider if a cut still feels too tight against a word.