Silence Remover

Tighten the pauses in a recording, or take the silence off the ends, in a single pass.

Runs on your deviceAudioWorkingNo account needed

What this tool does

Silence Remover measures the recording in 20-millisecond windows, takes the RMS level of each window across every channel at once, and marks anything below the threshold you set as silence. Runs of quiet windows at least as long as the shortest gap you allow are dropped and the surviving stretches are joined back together. Because the measurement is per window, cut points are quantised to those 20 ms boundaries — offset by whatever room tone you keep — rather than placed on an exact sample.

Two settings decide almost everything about the result. The threshold, from -70 to -20 dBFS with -45 as the default, has to sit above your noise floor and below your quietest wanted sound, and there is no automatic detection of where that is. Room tone to keep then decides how the joins sound: it holds back that much silence on each side of a cut, so a qualifying gap does not disappear, it collapses to exactly twice that figure. At the defaults, a four-second pause and a one-second pause both come back as 0.2 seconds, and an interior gap under half a second stays precisely as recorded. The two ends are the exception: even in the every-gap mode, whatever silence sits before the first sound and after the last is always cut back to the room tone, however brief it was, so the shortest-gap figure only ever governs the middle of the file.

There are limits worth knowing before you rely on it. The joins are hard cuts with no crossfade, so a steady background hides them and a changing one does not — a chair moving or a fan starting mid-file shows up as a step at the seam. It removes time rather than noise: a hiss or hum sitting above the threshold reads as sound, so nothing gets cut there, and no part of this attempts noise reduction. There is no waveform and no preview either, so you set numbers, read what the run reports, and adjust. It also cannot tell a deliberate dramatic pause from dead air; both are just quiet windows.

The work happens in this browser tab. The file is decoded by the browser's own audio decoder — or by a pure-JavaScript RIFF reader when it is a WAV — edited as raw samples, and written back out, so no copy of the recording is sent to a server. Writing MP3 is the one part that reaches out: the LAME encoder is a script fetched from this site the first time you ask for MP3, and nothing caches it for offline use, so a tab with no network can still produce WAV but not MP3. Two further things follow from doing the work here. The whole file is held in memory as 32-bit floats, roughly 20 MB for every minute of 44.1 kHz stereo, which is nothing for a voice memo and heavy for a ninety-minute interview. And the output is WAV or MP3 only, because browsers expose no AAC, FLAC or OGG encoder, so a FLAC or an M4A comes back as a WAV.

How to use it

  1. Choose the recording. MP3, WAV, M4A, AAC, FLAC, OGG and Opus are accepted, decoded by whatever your browser can play. Add several files and each one is processed separately under the same settings.
  2. Decide what to remove. Every silent gap tightens pauses throughout the file. Only the leading and trailing silence trims the head and the tail and leaves the interior timing exactly as it was, which also hides the shortest-gap setting because it no longer applies.
  3. Set the threshold against your noise floor. Start at -45 dBFS. Move it towards -20 when a room tone is stopping anything being cut, and towards -70 when quiet speech is being eaten. A file where no window at all rises above the threshold is refused rather than returned empty.
  4. Set the gap length and the room tone. Shortest gap to cut ignores anything briefer and accepts 0.05 s to 10 s; Room tone to keep, 0 to 2 s, sets how much silence survives on each side of a join. Output format starts on Match the input where possible, which returns an MP3 as MP3 and everything else as WAV, with the MP3 bitrate and the WAV bit depth underneath it — the file takes the original stem with -tightened on the end.

Processing & privacy

Runs on your device. Your file is processed directly in this browser.

Common use cases

Tightening a podcast interview

Thinking pauses and false starts add minutes to an episode. Cutting gaps over half a second shortens it noticeably while leaving the speech itself at its natural pace.

Cleaning up a voice memo

Recordings begin with the reach for the button and end the same way. Only the leading and trailing silence takes both ends off and leaves everything between them untouched.

Shrinking a lecture or meeting recording

Long silences while someone writes on a board are file size and listening time with nothing in them, and the run reports exactly how much of it went.

Preparing narration before an edit

Trim silence out of each take first, so the gaps you place afterwards in an editor are the ones you chose rather than whatever the room gave you.

Questions

Is my recording uploaded?
No. The samples are decoded, measured and written back out on your own machine, and no copy of the audio is sent anywhere — which is also why a long file depends on your device's memory rather than on a queue somewhere.
Why did nothing get cut?
Almost always a noise floor above the threshold. If a room hum sits at -35 dBFS and the threshold is at -45, every window counts as sound and there is no gap to find. Raise the threshold towards -20 until the quiet stretches fall below it, accepting that very quiet speech may go with them.
What exactly happens to a pause?
It is replaced by twice the Room tone to keep figure — at the default 0.1 s, every gap long enough to qualify comes back as 0.2 seconds however long it started out. Inside the recording, a gap shorter than the minimum you set, or shorter than twice the room tone, is left exactly alone; the silence at the very start and the very end is not measured against either figure and is always shortened to the room tone.
Will the cuts be audible?
Sometimes. There is no crossfade at a join, so a constant background covers the splice and a shifting one does not. More room tone usually helps; an obvious click at one particular join generally means the level either side of it was genuinely different.
Can I see where the cuts will land first?
No. There is no waveform and no preview. Afterwards you get the old and new durations, how much silence went, and — in the every-gap mode — the number of cuts, so the working method is to run it, read that line, and change the threshold if it looks wrong.
Does it change the quality of what is kept?
The kept stretches are copied sample for sample, with no resampling or filtering at the edit itself. The loss is all in the writing: MP3 in and MP3 out means a second lossy encode by the LAME encoder on this page, and MP3 output additionally pulls the file to a rate that format allows — 96 kHz becomes 48 kHz — and folds more than two channels down to stereo. WAV output avoids every one of those, at the cost of a much larger file.
Can I use it on a video's soundtrack?
It accepts audio files only, and the deeper problem is that it makes the file shorter. Audio pulled out of a video and tightened here will no longer match the picture, and any subtitle or chapter timings you already had drift with it.

Related tools