Skip to main content

Remove silence from audio online

Detect and cut silent sections with visual waveform, threshold control, and A/B preview.

Free & unlimited
All processing happens in your browser. No data is sent to any server.

This tool finds every stretch of the recording that stays below a loudness threshold for longer than a minimum duration, then splices the remaining audio back together sample by sample. It runs entirely on the Web Audio API: the file is decoded to a raw buffer, scanned in 50 ms blocks to draw the silence map you see before processing, then rebuilt into a new AudioBuffer holding only the kept regions. Four controls shape the result — threshold in dB, minimum silence length, padding kept on both sides of each kept region so words are not clipped, and an optional partial pause you can leave behind so speech does not sound machine-gunned. Five presets cover podcasts, music gaps, lectures and aggressive trimming, and the stats panel reports exactly how many seconds and what percentage came out.

Key facts about Remove silence from audio online

Key facts about Remove silence from audio online
MechanismSample-level splice in the browser: decodeAudioData, threshold scan, then a rebuilt AudioBuffer of the kept regions
Podcast preset-30 dB threshold, 0.8 s minimum silence, 0.1 s padding, 0.15 s of pause left in
Music preset-45 dB threshold, 2.0 s minimum silence, 0.2 s padding, no pause kept
Aggressive preset-25 dB threshold, 0.3 s minimum silence, 0.05 s padding, no pause kept
Gentle preset-40 dB threshold, 1.5 s minimum silence, 0.15 s padding, 0.3 s of pause left in
Lecture preset-35 dB threshold, 1.0 s minimum silence, 0.1 s padding, 0.2 s of pause left in
Analysis resolutionThe silence map is built from 50 ms blocks; region boundaries are found at single-sample resolution
Detection channelThreshold detection reads channel 1 only, but every channel is spliced identically so stereo stays in sync
PaddingKept regions are extended outward by the padding value so the first consonant and the last breath survive the cut
Output16-bit PCM WAV with a -no-silence suffix, at the source's own sample rate and channel count
StatsReports original length, new length, seconds removed, percentage removed and the number of regions kept
Accepted inputMP3, WAV, OGG, FLAC, AAC and M4A, up to 500 MB

What happens to your file

Nothing is uploaded and no engine is downloaded. Your file is read into an ArrayBuffer inside the page, decoded by the browser's own decodeAudioData, analysed and spliced by JavaScript running in this tab, and written back out as a WAV whose 44-byte RIFF header this page constructs by hand. There is no server call in the whole path — open the Network panel and process a file to confirm it stays empty. The result lives as a blob: URL in your tab's memory until you download it, and is released when you close or reload the page.

About this tool

  1. 1

    Drop the recording in

    MP3, WAV, OGG, FLAC, AAC or M4A up to 500 MB. The file is decoded and the silence map appears immediately, with detected silent regions marked along the timeline.

  2. 2

    Start from a preset

    Podcast for speech with natural pauses, Lecture for long dead air, Music for gaps between tracks, Gentle when you only want to lose the really long silences, Aggressive when you want everything tightened.

  3. 3

    Adjust the threshold

    Anything quieter than this counts as silence. Raise it toward -25 dB if room tone is being kept as audio; lower it toward -45 dB if breaths and soft words are being cut out.

  4. 4

    Set the minimum length

    Only gaps longer than this get removed. Below about 0.3 s you start cutting the natural spaces between words, which is what makes edited speech sound unnatural.

  5. 5

    Add padding and a partial pause

    Padding keeps a slice of audio either side of each kept region so consonants are not clipped. The partial-pause control leaves a fraction of each removed gap in place so the result still breathes.

  6. 6

    Process and check the stats

    The panel reports seconds removed and the percentage cut. If more than about a third of a speech recording vanished, your threshold is too high — lower it and run again before downloading.

Specs & compatibility
Input formatsMP3, WAV, OGG, FLAC, AAC, M4A — anything decodeAudioData can open
Maximum file size500 MB per file
Output format16-bit PCM WAV
Sample rate and channelsPreserved exactly; only the length changes
Threshold rangePresets span -25 dB to -45 dB, with a manual control for anything between
Minimum silence0.3 s to 2.0 s across the presets, adjustable by hand
BrowsersChrome, Edge, Firefox and Safari 14.1+ — AudioContext and AudioBuffer only, no WebAssembly
SpeedThe scan and splice are straight array work: a one-hour recording processes in a few seconds once decoded
MemorySource and result are both held as 32-bit float buffers, so peak RAM is roughly twice the decoded size
  • Watch the percentage removed. On a normally paced podcast, 10-20 percent is a healthy trim; 40 percent means the threshold is eating quiet speech.
  • Padding is the single control that decides whether the result sounds edited or broken. Below 0.05 s you will hear consonants get clipped off the front of words.
  • Leave a partial pause in for anything with a human speaking. Speech with every gap removed sounds breathless and is genuinely harder to follow.
  • For music, raise the minimum silence to two seconds so you only remove the gaps between tracks and not the quiet passages inside one.
  • If the recording has a constant hum, clean it with the noise reducer first — a high noise floor makes silence detection fail because nothing ever drops below the threshold.
  • Detection reads the first channel only. On a two-person interview recorded to separate left and right channels, run each side separately or mix to mono first.
  • The output is WAV, so if you are publishing an MP3, encode it once afterwards with the audio converter rather than round-tripping.
  • Visual silence map drawn before you process anything
  • Five presets covering podcast, lecture, music, gentle and aggressive trimming
  • Adjustable threshold, minimum silence length and padding
  • Partial-pause control so the result keeps natural rhythm
  • Before and after playback of the same file
  • Stats panel: seconds removed, percentage removed, regions kept
  • Sample-accurate splicing with stereo channels kept in sync
  • Tightening a podcast interview where both people left long gaps between answers
  • Cutting dead air from the beginning and middle of a lecture or webinar recording
  • Removing the silence between tracks in a single long recording of a vinyl side
  • Shortening a voice memo before sending it
  • Cleaning up a screen-recording narration where you paused to read your notes
  • Compressing hours of field recording down to only the parts with actual sound
They solve the same problem with different engines and different controls. This page does the whole job in the Web Audio API: it decodes the file, finds silence at sample resolution, and splices a new buffer, which lets it show you a silence map first, apply padding on both sides of each kept region, and leave a partial pause behind. The pause remover runs ffmpeg's silenceremove filter instead, which is faster on very long files and returns the file in its original format rather than WAV. If you want to see and tune what will be cut, use this one.
Almost always because the threshold is too high for the recording. The threshold is an absolute level, not a relative one, so a quietly recorded voice can sit below a -30 dB threshold and be treated as silence. Lower it toward -40 or -45 dB and the quiet passages will be kept. The other cause is padding set too low: at 0.05 s the splice happens very close to where speech restarts, so soft consonants at the start of a word can be clipped. Raise padding to 0.1-0.2 s for speech.
Because every pause was removed and nothing was put back. Human speech has rhythm; when every gap longer than the minimum disappears entirely, sentences run into each other and the result is exhausting to listen to. Use the partial-pause control to leave 0.15-0.3 s of each gap behind — that is what the podcast, lecture and gentle presets do. Raising the minimum silence length also helps, because short thinking pauses stay untouched and only genuinely long dead air is removed.
Yes. Silence detection reads the first channel to decide where the cuts go, but the splice is then applied to every channel using the identical sample ranges, so left and right stay aligned to the sample. The one case to watch is a recording where the two channels hold different sources, such as an interview with one microphone panned hard left and the other hard right. There, channel one alone decides what counts as silence, and the second speaker's audio can be cut when the first is quiet.
16-bit PCM WAV, at the source's own sample rate and channel count, named with a -no-silence suffix. WAV is used because it is the only lossless format the browser can write without loading an external encoder, which means this step adds no compression artefacts of its own. It is also considerably larger than a compressed source — roughly 10 MB per stereo minute at 44.1 kHz. If you need MP3 or AAC for publishing, run the WAV through the audio converter afterwards so the lossy encode happens once.
No. This tool uses nothing but the Web Audio API built into your browser. The file is read into memory in the tab, decoded by decodeAudioData, scanned and spliced by JavaScript on this page, and written back to a WAV by hand-built RIFF header code. It does not even download a processing engine — unlike the converter tools here, there is no ffmpeg fetch, so after the page loads there is no network activity at all. The result is a blob: URL in your own tab and vanishes when you close it.

Related tools

View all

Part of Audio tools that run in your browser

Updated

We use anonymous analytics to improve ToolChamp. No personal data is stored or sold. Privacy Policy