ByteScope

Silence Remover

Drop a recording — or a video, its audio track is used — and see the pauses shaded on a waveform. Set the threshold and minimum silence length, click any cut to keep it, play the result before exporting, then download a WAV or an EDL, FCPXML or CSV cut list for your editor. The file never leaves your computer.

Your files never leave your browser — all processing is local.

Drop an audio or video file, or click to choose

One file at a time: MP3, M4A, WAV, OGG, FLAC, or a video's sound track

Before you start

  • It detects volume, not meaning. Loud music or an air conditioner can hide silence, and breaths, "um" and "uh" are not removed.
  • Cutting pauses makes speech sound faster. Very small pads sound hurried; the defaults are cautious.
  • For video this version only writes cut lists (EDL, FCPXML, CSV). The cutting itself happens in your editor.
  • Importing the cut lists into DaVinci Resolve, Final Cut Pro and Premiere Pro is untested so far. Check the first import against the preview.
  • The whole file is decoded into memory: about 93 minutes mono or about 1 hour stereo at 48 kHz. The page tells you the exact limit for your file.
  • Your file is processed in this tab and never uploaded.

About this tool

This page finds the silences in a recording and cuts them out. Detection is by volume, not by words: the file is split into short blocks, each block's level is measured as RMS in dB with any DC offset removed, and a stretch counts as silence while it stays below the threshold. The threshold can be automatic — about 10 dB above the quietest tenth of the recording, kept between -70 and -30 dB — or set by hand from -70 to -20 dB. Only silences at least as long as the minimum (600 ms by default, adjustable from 200 to 3000 ms) are ever cut, and every kept stretch gets padding, 100 ms before and 150 ms after by default, so words do not start or end abruptly. Nothing is uploaded: decoding, detection and export all happen in your browser, and from a video only the audio track is read.

Preview before you export

The timeline draws the waveform with the cuts shaded, and a summary line shows the effect, such as 40:12 → 31:05 with the number of cuts. Click a shaded cut to keep it after all. Moving a control re-plans the cuts and clears those manual keeps, and the page says so. Play runs the cut version straight from the decoded audio, jumping over the cuts with short fades, so you hear the result before anything is written. The new length shown is calculated the same way the export does it, so it matches the file you get.

What you can export

For audio, a WAV at the source's sample rate, with a 10 ms equal-power crossfade at every join to avoid clicks. For video, this first version writes cut lists rather than a rendered video: an EDL (CMX 3600), an FCPXML 1.9 file and a CSV, meant for DaVinci Resolve, Final Cut Pro and Premiere Pro. Pick the frame rate, from 23.976 to 60 with drop-frame choices for 29.97 and 59.94; it is prefilled from the file. Import the media into your editor first, then the cut list. The FCPXML was checked against Apple's DTD, but import into Resolve, Final Cut Pro and Premiere has not been tested in those programs yet, so treat the first import as a trial and check the cut points. A ready-cut video file is not offered yet.

What it will not do

It detects volume, not meaning. Loud background music or an air conditioner can hide silence, and breaths, um and uh are not removed: taking out filler words needs word-level speech recognition, which this page does not use. Cutting pauses also speeds up the pacing, and with very small padding speech sounds hurried, so the defaults err on the cautious side. For a cleaner result, run Audio Denoise first and Loudness Normalizer afterwards.

Frequently asked questions

Is my recording uploaded anywhere?

No. The file is decoded by your browser, analysed by code running in the same tab, and the WAV or cut list is written there too. Nothing is sent to a server, because there is none behind the page, and nothing about the file — its name, length or content — leaves the browser.

Does it remove um, uh and breaths?

No. It detects how loud each moment is, not what is being said, and a breath or an um is usually louder than a real pause, so it stays. Removing filler words needs speech recognition with word-level timestamps, which this page does not have. What it does remove is the empty gaps between sentences, which is where most of the time goes in a lecture or a solo recording.

Why is some silence not detected?

Because it is not quiet enough for the threshold. Background music, an air conditioner or a noisy room keep the level above the line even when nobody speaks. Raise the manual threshold toward -30 dB, or try a recording cleaned with Audio Denoise first. Note too that only silences at least as long as the minimum (600 ms by default) are cut, so short pauses between words are left as they are.

Why does the result sound rushed or choppy?

Because too much was cut or the padding is too small. Each kept stretch gets 100 ms before and 150 ms after by default; speech sounds natural with some air around it, so raise the padding or the minimum silence length if the result feels hurried. A cut taken out of the middle of a breath can click, which is why audio exports add a 10 ms crossfade at every join.

Can I get a video with the silences cut, and will it import into my editor?

Not a finished video yet: for video files the page writes an EDL (CMX 3600), an FCPXML 1.9 file and a CSV, which you open in DaVinci Resolve, Final Cut Pro or Premiere Pro to do the cutting. The FCPXML passed Apple's DTD check, but importing it into those programs has not been tested yet, so check the first import against the preview. Bring the media into the editor before importing the list.

How long a file can I use?

About 93 minutes of mono or about 1 hour of stereo at 48 kHz. The browser decodes the whole file into memory, and Chrome's decoder stops at about 268 million samples per channel. The page checks the length before it starts and tells you the exact limit for your file. For something longer, cut the recording into parts first.

Related tools