Remove silent gaps from podcasts and voice recordings

How to use the silence remover

1. Add the recording (MP3, M4A, WAV, OGG, FLAC, WebM, or a video’s audio).
2. Set Silence below, from −80 to −10 dBFS. The default −45 suits a quiet room.
3. Set Shortest gap to remove (default 0.6 s). Shorter pauses are left alone.
4. Set Room tone to keep at each cut (default 0.25 s).
5. Choose Save as and press Remove silence. The result lists Time saved, Gaps cut, Original length and the new Length.

What it does and when to use it

Long pauses creep into lectures, interviews, screen-recording voice-overs and voice notes. This tool tightens them in one pass without touching what was said, a quick first edit before you publish a podcast. It does not detect “um” and “ah”; those are speech, not silence.

After cutting, level the result with the audio normaliser. To also play it faster, the audio speed changer can speed up speech while keeping the voice’s pitch. For a single unwanted section, the audio trimmer is simpler.

How it works

The decoded audio is split into 10 ms windows, and the RMS level of each window is compared with the threshold. A run of quiet windows counts as a gap if it lasts at least the shortest-gap value (or the room-tone value, if that is larger). For each gap in the middle of the file, half of the room tone is kept on each side and the rest is cut. The kept pieces are joined with a 5 ms crossfade so the edits do not click. Everything runs in a Web Worker on your device; nothing is uploaded.

Worked examples

A test file of 18.9 s: 4 s of sound, a 3 s gap, 5 s of sound, a 1.5 s gap, 3 s of sound, a 0.4 s gap, then 2 s of sound.

• Defaults (−45 dBFS, 0.6 s, 0.25 s): Gaps cut 2, Time saved 0:04.01, Length 0:14.89. The 0.4 s pause is shorter than 0.6 s, so it stays.
• Shortest gap 2 s, room tone 0.5 s: only the 3 s gap qualifies. Gaps cut 1, Time saved 0:02.51, Length 0:16.39.
• A file with no long pauses returns a note: “No gaps longer than 0.6 s below -45 dBFS were found”, with a hint to raise the threshold for noisy recordings.

Limits and tips

• Music with quiet passages can be damaged. The tool is built for speech.
• Music beds under speech fill the gaps, so nothing is found; remove silence before adding music.
• The whole file is decoded into memory. A one-hour stereo podcast needs well over 1 GB, which phones may not have; split very long files or use a desktop browser.
• Non-WAV files use the browser’s WebCodecs decoder. MP3 output and rare inputs (AIFF, AMR, WMA, CAF) need the optional ffmpeg fallback, a one-time download of about 30 MB from cdn.jsdelivr.net.