Silence Remover
Remove silence from audio automatically. Two numbers decide what goes: how quiet a gap has to be, and how long it has to last.
Runs entirely in your browser — your files and text never leave your device.
Drag & drop your file here, or
Trims leading and trailing silence and long gaps that fall below the threshold.
M4A (AAC) and Opus are compact and play everywhere; WAV is uncompressed and FLAC is compressed lossless — both larger, best for editing. MP3 export isn't offered (see the FAQ).
Working on your device — keep this tab open until it finishes.
How to use
- Put the recording in the Audio file box. Whatever your browser can decode is fair game, including MP3 and M4A, and a player turns up underneath once the file has been read.
- Silence threshold (dB) sets the level below which audio counts as nothing. It opens at -50, the box runs from -100 up to 0, and it moves a whole decibel at a time.
- Min silence (s) is how long a quiet run has to last before it is worth removing. 0.3 is the default and 0.1 is the step. Push it to 0.6 or 0.8 to protect breaths and thinking pauses.
- Choose an output format. WAV and FLAC keep the surviving samples exactly as they were, while M4A and Opus re-encode them much smaller.
- Press Remove silence & download and let the bar finish. Play the result before saving: speech sounding clipped means the threshold is too high, and a file the same length as before means it is too low.
About this tool
Dead air at both ends, and a two-second think before every answer, will pad a recording out by minutes. To remove silence from audio without opening an editor, you only have to answer two questions: how quiet does a stretch have to be before it counts as nothing, and how long does it have to run before it is worth cutting. The two boxes on this page are those questions.
The pass walks the file sample by sample and marks everything under the threshold as quiet. At the default of -50 dB that bar sits at an amplitude of about 0.003. A marked run only counts once it lasts at least as long as Min silence, which starts at 0.3 seconds. Every run that qualifies is then cut apart from 50 milliseconds left at each end, so a removed pause comes back as a tenth of a second of breathing space rather than a hard splice. Leading and trailing silence goes the same way.
Voice memos, lecture recordings, interviews where the other person thinks before answering: those are the files that shrink most. Exports are WAV, FLAC, M4A or Opus. The lossless pair hands the surviving samples over untouched, with WAV written as 16-bit PCM. M4A and Opus squeeze the file down small enough to email.
When a pass removes nothing at all, the room was louder than you thought and the threshold has to come up. A fan, a laptop, traffic through a window: all of that sits above -50. Raise the number 5 dB at a time and listen. If the hiss is bad enough that speech and silence read alike, clean the file with the noise reducer first, then come back here. Lifting one specific section out by hand is a job for the Audio Cutter, and the Audio Normalizer evens out whatever survives.
Trimming an interview that breathes too much
Say you have a 12-minute interview where the person answering takes about two seconds to gather their thoughts each time. Leave the threshold at -50, set Min silence to 0.6 so breaths and short pauses survive, and run it. Each of those two-second gaps comes back as 0.1 seconds, since 50 milliseconds is kept at both ends of every cut. Thirty of them is 57 seconds off the running time.
If that first pass barely dents the length, the room tone is sitting above your threshold. Move it to -45, run it again, and listen to the result player before you press Download. Two passes at different settings usually lands it.
Good to know
- Detection reads the instant level of each sample with no averaging over a window, and that has two consequences worth knowing. A single click, a chair creak or one loud breath in the middle of a pause splits that pause into two shorter runs, and if neither half reaches your minimum length the gap survives untouched. The tool also has no idea which silences carry meaning. A dramatic beat before a punchline reads exactly like dead air, so play the result through once before you publish it.
Frequently asked questions
What actually counts as silence?
Anything quieter than the threshold you set, running for at least the minimum you set. Those are the only two tests. At -50 dB the bar sits at an amplitude of roughly 0.003, quiet enough that little except near-digital silence gets under it. Move the number toward -40 and ordinary room tone starts counting as silence too.
Why did it trim the ends but leave the middle alone?
The head and tail of a recording are usually properly silent, because nothing was happening before you started talking. Pauses in the middle carry room tone, and room tone sits above -50 in most spaces. Raising the threshold to -45 or -40 brings those pauses into range. Keep an ear on quiet speech as you climb, since that is what starts vanishing around -30.
Can you hear where the cuts were made?
Rarely, and the 50 milliseconds kept at each end of a cut is why. Both sides of a join are already quiet, so the seam falls between two near-silent samples instead of slamming one word into the next. A generous minimum length helps as well: cut only the long pauses and the rhythm of the speech survives.
Will it keep the breaths between sentences?
Set Min silence above the length of a breath and they stay. A breath is usually short enough that 0.6 protects it while still clearing a three-second stretch of nothing. Dropping the minimum toward 0.1 is what makes speech sound airless and rushed.
Where is the MP3 option?
There isn't one. An MP3 encoder needs a licence that isn't bundled here, so the row offers M4A, WAV, Opus or FLAC. M4A is the safe pick of those four: it plays on phones and laptops without anyone thinking about it, and at matching quality it is smaller than MP3 would have been.
Does the audio that survives get changed at all?
The kept segments are copied out sample for sample. No gain is applied, no fades are drawn in, and the channel layout and sample rate carry straight over from the source. WAV comes back as 16-bit PCM, and the compressed formats re-encode, which is the only place any quality goes.
Missing something in Silence Remover? Suggest a feature →