Remove silence from audio

The tool measures the loudness of the recording in 10 millisecond slices, in decibels (dB). Anything that stays under the threshold for long enough is treated as silence. Silence is never truly empty: it holds the background noise of the room, the microphone or the street. So the right threshold sits just above that noise, and well below the voice.

The automatic threshold starts from your file. The tool measures its noise floor and the level of the speech, then places the threshold between the two, closer to the noise. It is drawn as a dashed line on the waveform, so you can see what falls under it. The "Sensitivity" slider lets you move it, and the tool tells you whether your setting is more careful or more sensitive than its suggestion.

Two safeguards stop sentences from being chopped up. The voice has to drop 3 dB below the threshold before it counts as finished, which protects the tails of words. And a very short sound that is barely louder than the threshold (a click, a lip noise) does not split a silence in two.

• Low threshold (around −60 dB): only true dead air is cut.
• High threshold (around −30 dB): breaths and soft sentence endings go too.
• "Ignore silences shorter than": the small gaps between words stay where they are.
• "Keep around speech": the margin left on each side of a sentence.

The four presets all start from the automatic threshold and mostly change the pacing: how long a silence must be before it is touched, how much of it is left, and the margin kept around speech. Each one can be adjusted afterwards, slider by slider.

For a podcast or an interview, pauses are shortened rather than erased, so the conversation still breathes. A voice note is meant to be quick, so silences can go entirely. In a lecture or a talk, short pauses are part of the delivery: only the long gaps (the speaker writing on the board, the room waiting) are reduced. "Aggressive" tightens everything, at the cost of a less natural flow.

• Podcast: silences of 0.6 s or more, shortened to 0.35 s, 0.12 s margin.
• Voice note: silences of 0.4 s or more, removed, 0.1 s margin.
• Lecture / talk: silences of 1.5 s or more, shortened to 0.6 s, 0.25 s margin, threshold 3 dB more careful.
• Aggressive: silences of 0.2 s or more, removed, 0.05 s margin, threshold 4 dB higher.

"Remove" takes each silence out completely, apart from the margin around speech. It saves the most time and suits dictation, voice notes and voice-over recorded in several takes. The drawback shows on real conversation: sentences run into each other with no room to breathe.

"Shorten" brings every silence down to a fixed length, 0.35 seconds for example. A pause carries meaning: the end of a thought, a hesitation, a deliberate beat. Reducing it instead of deleting it keeps the punctuation of speech and only removes the waiting. This is the setting to use for a podcast.

Either way, every splice gets a 5 millisecond crossfade between the sound before and the sound after, so there are no clicks. And you stay in control: tap a red area on the waveform, or "Keep" in the list of cuts, and that silence stays in.

MP3 plays on every device and in every messaging app. For speech, 128 kbit/s is plenty; 64 or 96 kbit/s makes a smaller file for a voice note, and 192 kbit/s or more is for recordings with music.

M4A (AAC audio) is smaller than MP3 at the same quality. It plays on phones, in browsers and in editing software. It is only offered when your browser has an AAC encoder.

WAV is uncompressed: no loss at all, but roughly 5 MB per minute in mono and 10 MB in stereo. Choose it if you will keep editing in another program. The tool shows the estimated size before you export and flags a file that is larger than the original.

One caveat: if your file is already an MP3 or an M4A, exporting to a compressed format encodes it a second time. At 128 kbit/s and above, the difference is not audible on speech.

Most online tools upload your recording to their servers. This one does not: the program is loaded into your browser, then it works on your phone or computer. An interview, a lecture or a private message stays with you, and there is nothing to delete from a server afterwards.

The file is never loaded into memory in one piece. It is decoded a little at a time, measured, then written out as it goes, which is how recordings of up to 3 hours can be handled. Speed depends on your device, not your connection. In our tests on a recent laptop, a two and a half hour MP3 was analysed in about twenty seconds and exported in about the same time. A phone is slower.

Your settings are remembered in the browser for next time. Your recordings never are.