1. Remove Silence
  2. From audio

Remove silence from audio

Podcasts, voice memos, lectures. The cut happens on your device.

Spoken sample, 5 pauses

Sample shown. Choose a file to clean your own.

KeptCutDrop an audio or video file here
Before
17.0 s
After
9.0 s
Cut
8.0 s 47%
Detection settings

How to remove silence from an audio file

  1. Choose an MP3, WAV, M4A, AAC or FLAC file, or drop it on the waveform.
  2. Check the shaded gaps. Raise Silence below if quiet room noise is left in, lower it if soft words are cut.
  3. Play the result and listen to the joins.
  4. Download the WAV and use it in your editor or podcast host.

The page measures loudness, so it works on any language and any voice. It does not transcribe anything, and it does not remove filler words such as “um”. For those, see how to remove filler words.

What each setting does

There are three sliders. All three update the waveform as you drag, so you can see the effect before you play anything.

SettingRangeStarts atWhat it changes
Silence below−70 to −20 dBSet from your fileAnything quieter than this level counts as silence.
Shortest pause200 to 2,000 ms500 msQuiet stretches shorter than this are left alone.
Keep around speech0 to 400 ms120 msSilence kept before and after each word, so starts and ends are not clipped.

When you open a file, the page reads its quietest and loudest parts and places Silence below between them, somewhere from −60 to −24 dB. That guess suits most clean speech. Move it when the shaded gaps do not match what you hear.

The before, after and cut figures above the sliders show how much time you save. If the cut looks large for the kind of recording, play it back and check that words are not being taken with the pauses.

Settings that suit common recordings

RecordingShortest pauseKeep around speechResult
Podcast or interview600 ms150 msConversation keeps its rhythm.
Voice memo or lecture400 ms100 msLong thinking pauses go, sentences stay intact.
Tight voice-over250 ms60 msEvery gap closes.

For a deeper look at choosing a threshold for noisy rooms, quiet speakers and music beds, read silence threshold settings.

How the detection works

The page splits the recording into 10 ms slices and measures the loudness of each one. A run of slices below Silence below that lasts at least the Shortest pause becomes a gap. The page then gives back Keep around speech on both sides of each gap, so the words keep their natural attack and release, and joins the remaining parts end to end.

Because it only looks at level, it treats any steady quiet sound as silence and any loud sound as speech. A cough, a door or a keyboard click in a pause is kept. A whispered aside below the threshold is cut. That is why it pays to listen to the result rather than trust the figures.

Silence at the start and end

Dead air before the first word and after the last is cut with the rest, and it is cut completely: Keep around speech only pads the joins between words. If you mostly want the ends trimmed and the pauses inside left alone, raise Shortest pause to 2,000 ms. Then only gaps longer than 2 seconds are cut, wherever they are.

Common mistakes and how to check the result

  • Threshold too high. Soft word endings and quiet consonants fall below it and get cut. Listen for words that sound clipped at the end, then lower Silence below by 3 dB.
  • Threshold too low. Room hum, fans and traffic sit above it, so nothing is found. Play a quiet stretch, watch the waveform, and raise the threshold just above it.
  • No padding. With Keep around speech at 0, breaths and word onsets are trimmed hard. Keep at least 60 ms unless you want a deliberately tight cut.
  • Music under the voice. A music bed fills the pauses, so the level never drops. Cut the voice track before you add music.
  • Two people on one file. A quiet guest can fall below a threshold set for a loud host. Set it from the quieter voice.

To check a cut, play the result from the start and listen at normal speed. Pay attention to the first second after each join. A good cut sounds like the speaker simply talked a little faster.

Files that work

The page uses your browser’s own decoder. MP3, WAV and M4A open in every current browser, and most AAC, FLAC, MP4 and MOV files do too. If a file will not open, the page says so instead of guessing. Files over 400 MB are too large for a browser tab to hold in memory.

The download is a 16-bit WAV at the sample rate of your file, with the same number of channels. Import it into your editor, your podcast host or a converter as you would any WAV.

Why nothing is uploaded

Your browser decodes the file and the page measures its loudness every 10 milliseconds. The cut and the WAV are built in this tab, so a private recording stays private.

Before and after you cut

  • Cut before you process. Remove silence before you add compression, noise reduction or music. Those raise the level of the pauses and make them harder to find.
  • Keep the original. The page never changes your file, so the original stays where it was. Name the download so you can tell the two apart.
  • Normalize last. Shorter pauses do not change how loud the speech is. If you publish a podcast, set loudness after the cut.

When to use something else

You need toUse
Export a recording longer than 3 minutesAudacity’s Truncate Silence, free on Mac, Windows and Linux
Cut the video, not only its soundRemove silence from video
Remove um and uhRemove filler words
Clean a voice memo on your phoneVoice Memos on iPhone
Even out loudness for a podcastPodcast loudness and LUFS

On iPhone and iPad, QuellCut does the same detection on recordings of any length up to 24 hours, adds filler word and repeated-take removal, balances loudness to −16 LUFS and exports M4A or WAV. Like this page, it works on the device.

Questions

Is the browser tool really free?
Yes. You can preview a recording of any length, and downloading the cleaned WAV is free for recordings up to 3 minutes. There is no account and no watermark.
Is my audio uploaded anywhere?
No. Your browser decodes the file and builds the result in the open tab. Nothing is sent to a server, so you can use it for private interviews and voice notes.
Why do I get a WAV and not an MP3?
WAV is uncompressed, so cutting does not add a second round of compression. Every editor and most podcast hosts accept it, and you can convert it to MP3 or M4A afterward if you need a smaller file.
Can it remove um and uh?
No. Filler words are as loud as the words around them, so a level meter cannot find them. That needs speech recognition, which is covered in the filler words guide.
What if my recording is longer than 3 minutes?
You can still preview the cut here. To export a longer file, use an editor such as Audacity, or QuellCut on iPhone and iPad, which handles recordings up to 24 hours.