Cut the dead air, keep the pauses
Drop a recording. The tool finds every stretch quieter than −40 dB that lasts longer than half a second, shows you where it is, and cuts it out with a 150 ms margin so no word loses its tail. One file in, a shorter one out, nothing uploaded.
The Silence Remover cuts the dead air out of a podcast, interview or lecture. You set the level below which audio counts as silence and the length of gap to remove, and the tool keeps a short margin around each cut. It shows how much shorter the file becomes.
How it works
Add the recording
Open the interview, lecture or voice note with the pauses in it.
Set what counts as silence
Choose how long a pause has to be before it is cut, and how much of it stays.
Download the shorter file
Save the trimmed audio and see how much time it saved.
The silence is not the problem
Listen to a good interview and you will hear pauses everywhere: before an answer, after a joke, in the second a person needs to find a word. Those pauses are rhythm. What makes a recording feel slow is a different kind of quiet - the three seconds after a microphone is muted, the eight seconds while somebody looks for a file, the minute at the end that nobody edited out.
A silence remover is good at the second kind and dangerous with the first. That is why the tool asks for a minimum length instead of cutting everything below a level: half a second of quiet is usually a thought, two seconds is usually a gap. The default, 0.6 seconds, sits between them.
The other number is the margin kept around each cut. Without it, the trailing breath of a word is the first thing to go, and the speaker starts to sound clipped - as if every sentence began and ended inside a cylinder. 150 milliseconds is short enough to remove the dead air and long enough to keep the consonant that was decaying when the room went quiet.
What the cut looks like
Two gaps, 0.9 and 2.4 seconds, both silent enough and both long enough to count. The recording loses 3.3 seconds and not one word. In a 20-minute lecture the same settings usually find 40 to 90 of these, which is where the ten minutes people complain about actually live.
Two gaps, 3.3 seconds, removed without touching a word.
How long a gap is worth cutting
| Recording | Usual dead air | Minimum gap to use | Why |
|---|---|---|---|
| Interview, two mics | 0.8–2 s between answers | 0.6 s | answers start after a breath |
| Lecture or screen recording | 2–8 s while switching slides | 1.2 s | long, obvious gaps |
| Voice note to a colleague | 0.5–1 s | 0.5 s | short file, little to gain |
| Podcast, edited later | 1–3 s | 0.8 s | keeps the banter, removes the fumbling |
| Meditation or ASMR | long on purpose | do not cut | silence is the product |
If a file has music or room tone under the voice, the threshold has to go lower than the music, not lower than the voice. Otherwise the quiet passages of the music get cut and the result sounds like a radio with a loose cable.
The two numbers, explained
The threshold is measured in dBFS, where 0 is the loudest a file can be and the numbers go negative from there. Normal speech peaks around −6 to −12 dB and averages −20; a quiet room sits near −55 to −65. −40 dB therefore means "quieter than a room with nobody talking", which is exactly what a gap is.
If the recording was made on a laptop fan or a noisy street, the room floor can be −35 dB. Raise the threshold to −30 and the tool will still find the gaps; leave it at −40 and it will find nothing, because nothing is ever that quiet.
Cutting exactly at the threshold clips the decay of the last consonant. "S" and "f" are mostly air; they fade rather than stop, and their tail is quieter than the threshold before the word is actually over. Keeping 150 ms on each side of the cut protects that tail and the intake of breath before the next sentence.
Raise it to 300 ms if the speaker is fast and the cuts sound abrupt, or if the recording is compressed and every consonant has a soft edge. Lower it to 50 ms only for files where the silence is genuinely empty - a digital mute, not a room.
Do not cut a file twice. Removing silence from an MP3 and saving it as MP3 encodes it again, and the second generation is audibly duller. If the result will be processed again, export WAV.
Do not use this on music. The quiet parts of a song are not gaps, and a threshold that finds them will cut the intro, the bridge and the ending.
And do not trust a percentage as a quality measure. A file that got 30 percent shorter is not automatically better; it is shorter. Listen to the first minute before you export the rest.
Signs you cut too much
- check_circleSentences start on a hard consonant with no breath in front of them
- check_circleThe speaker seems to interrupt themselves, though nobody spoke over them
- check_circleThe rhythm is even but the meaning is not - two ideas run together
- check_circleRoom tone disappears completely between words, making the audio feel sewn together
- check_circleA laugh, a sigh or a thinking sound has gone missing mid-answer
Questions about the cut
Does it remove background noise as well?expand_more
What file formats can I drop in?expand_more
Why does the result sound slightly different at the cuts?expand_more
How much shorter will my file get?expand_more
Is the audio uploaded to a server?expand_more
Can I hear the result before downloading?expand_more
A shorter file, same words
Pick the recording, check the red stretches on the waveform, cut and download.
fact_checkQuick facts
- The threshold is a level, not a length: everything quieter than it counts as silence, so a noisy room needs a higher threshold before anything is removed.
- Keeping about 0.2 seconds on each side of a cut preserves the natural rhythm of speech; cutting exactly on the waveform makes the result sound clipped.
- 44.1 kHz, 16 bit, stereo means 44,100 samples × 2 channels × 2 bytes each second: 1411 kbit/s, or about 10 MB per minute.menu_bookMicrosoft Learn
- WAV stores the samples as they are, so it is the right working format for editing and the wrong one for sending.menu_bookMicrosoft Learn
- MP3 is lossy: decoding never returns the waveform that was encoded, which is why repeated re-encoding lowers quality each time.menu_bookFraunhofer IIS
helpQuestions people ask
How do I remove pauses from a podcast or interview?expand_more
What threshold should I use to remove silence?expand_more
Will removing silence make a talk sound unnatural?expand_more
hubWhat this page is about
- WAV (RIFF)open_in_new
An uncompressed container from Microsoft and IBM built on RIFF; CD-quality stereo at 44.1 kHz and 16 bit needs 1411 kbit/s.
- MP3open_in_new
A lossy audio format from MPEG-1 Audio Layer III, developed at Fraunhofer IIS, that removes sounds the ear is unlikely to notice.
- Web Audio APIopen_in_new
The W3C API that builds an audio graph in the browser and can render it faster than real time with an OfflineAudioContext.
library_booksOfficial sources cited on this page
- Resource Interchange File Format (RIFF)Microsoft Learn
- MP3 — the audio coding format developed at Fraunhofer IISFraunhofer IIS
- Web Audio APIW3C
Continue with the same file
Each of these opens what the tool above produces. Use “Use in another tool” under the result to carry the file over — it moves from tab to tab inside your browser, never to a server.
Switched-off tools are not listed here: only what works right now.