How to Remove Silence from Audio Automatically
Updated 2026-07-15
To remove silence from a recording, open the Silence & Filler Word Remover, drop in your audio (or video — the tool reads its soundtrack), and it automatically finds every stretch of dead air. Review the detected cuts on the waveform, toggle off any you want to keep, and export the tightened audio as MP3 or WAV. The analysis runs entirely on your device — nothing is uploaded.
Why editors charge for this feature
Manually hunting pauses in an hour-long podcast means scrubbing the whole timeline; it's the single most tedious edit in spoken audio. Tools like Descript made "remove silence" a headline paid feature. The mechanics, though, are honest signal math: measure loudness over small windows, call anything below a threshold "silence" if it lasts long enough, and cut it out with tiny crossfades so the joins don't click. That's exactly what this tool does — locally.
Step by step
- Open the Silence Remover and drop in your file. The waveform renders with detected silences highlighted.
- Tune three sliders: threshold (how quiet counts as silence, default −40 dB), minimum silence length (default 0.6 s — protects natural pauses), and padding (a little air kept around speech so it doesn't sound clipped).
- Watch the summary update: "removes N gaps — saves M:SS". Every cut is listed; toggle off any pause that's rhetorically important.
- (Desktop, optional) Run the filler-word pass — an on-device Whisper transcription finds standalone ums, uhs and similar; each becomes a toggleable cut with the word shown, so you stay in control.
- A/B the original vs. the cleaned version with the preview buttons, then export MP3 or WAV. You can also download the cut list as CSV to reproduce the edit in your video editor.
Getting the threshold right
- Recorded in a quiet room? −45 to −40 dB catches everything cleanly.
- Noisy background (fans, street)? Raise it to −35 or −30 dB so room tone isn't mistaken for speech.
- Interviews with soft-spoken guests: lengthen minimum silence to 0.8–1.0 s so thinking pauses survive.
The right setting is the one where the summary count matches your ear — that's why every cut is previewable before anything is exported.
What about video jump-cuts?
This tool's exports are audio (plus the cut list). If your source is a video and you want the picture jump-cut to match, import the CSV cut list into your editor and razor at those times — or, for a talking-head where the audio matters most, replace the video's soundtrack with the cleaned audio. For trimming a video's start/end, the Video Trimmer does it without re-encoding.
Privacy
Podcasts routinely contain unreleased material, client names, and off-record moments. Cloud silence-removers upload all of it. Here the decoding, detection, filler transcription and encoding all happen in the tab — the recording never leaves your machine, and the optional AI pass downloads only the open-source model weights, never your audio. Pairs naturally with the Podcast Show Notes generator once the episode is clean.