Remove Silence From Audio

None — that is the point. Every recording has a different noise floor, so a number that works on one is wrong on the next, and "try −40 dB" is not advice. This measures the noise floor of your actual file and sets the threshold a fixed distance above it. The measured floor and the resulting threshold are both shown, and you can override with a fixed level if you would rather.

No. Everything runs in your browser, so nothing is transmitted and there is no file size limit. That matters more here than for most tools: this is the sort of thing you run on an unreleased episode or a confidential interview.

Because speech with no pauses sounds wrong. The pauses carry the phrasing — they are where a sentence ends and a thought turns — and removing all of them produces something that sounds hurried and obviously machine-edited, in a way listeners notice but cannot name. Shortening a two-second gap to half a second takes out the dead air and keeps the rhythm. Deleting them entirely is available as the jump-cut preset, which is the right choice for fast-paced video.

No, because every join is crossfaded. This is the single commonest failure of automatic silence removers: cutting a waveform at an arbitrary sample leaves a step discontinuity, and a few hundred of those is a recording full of ticks. Measured on a test signal, a hard cut leaves a jump about twenty-five times the natural sample-to-sample step; with the crossfade, it drops to well under twice — inaudible. The crossfade length is adjustable, and setting it to zero is measurably worse.

It is designed not to, in two ways. Speech opens and closes the gate at different levels, so the fading tail of a word cannot flicker it shut mid-syllable. And a configurable amount of silence is kept either side of every retained section, so attacks are not clipped. If you set the breathing room to zero and the threshold high, it will absolutely clip words — which is why the waveform shows you the cuts before you commit to them.

Then silence detection gets harder, because there is less difference between someone speaking and the room. The tool measures that gap and warns you when there is under about 12 dB between the noise floor and the loudest moment — below that, any threshold is a guess. The fix is to remove the noise first and then run this, which is a much cleaner result than any threshold can achieve on its own.

Yes — there is a preset for exactly that. Cutting the silence before you started talking and after you stopped, without touching anything in between, is one of the most common reasons people look for this.

It works on a video file’s soundtrack and gives you back audio. It does not currently cut the picture to match, so for video editing you would use the result as a guide rather than a finished track.

That depends entirely on the recording, which is why the tool tells you rather than promising a figure. A tightly delivered read might lose five per cent; an unrehearsed interview with long thinking pauses can lose twenty or more. If it reports removing over half the file, it says so as a warning rather than a triumph — that usually means the voice is quiet relative to the background and some speech may have gone with the silence.

M4A for something small that plays everywhere, WAV for the exact samples with no further compression, or Opus. M4A at 128 kbps is the sensible default for spoken word.

No. Nothing is compressed, levelled or filtered — sections are removed and the joins are crossfaded, and everything that remains is bit-for-bit what you gave it, apart from the few milliseconds inside each fade. If you also want the levels set, run the audio editor afterwards.