You record a one-hour interview and, reviewing it, notice whole stretches of silence: pauses to think, interruptions, moments where nobody's talking. Trimming all of that by hand, listening from start to finish, is the most tedious part of editing spoken audio. Removing silence automatically solves this, but the technical challenge is not wrecking the normal pauses of a real conversation.
The problem with "cut anything that's quiet"
A naive silence detection algorithm simply looks for stretches below a certain volume and deletes them. The problem is human speech is full of short pauses — between sentences, while breathing, while thinking of the next word — that are a natural part of how we talk. If the algorithm deletes any silence, no matter how short, the result sounds rushed and artificial, like someone talking without breathing.
The two parameters that solve this
Silence threshold (in dB): the volume below which something counts as silence. A value closer to 0 dB detects silence more easily (it's stricter about what counts as "sound"); a lower (more negative) value requires a stretch to be genuinely silent before removing it. -35 dB is a reasonable starting point for normal voice recordings; noisier environments may need adjusting.
Minimum duration (in seconds): only silence lasting at least this long gets removed. This is the parameter that prevents wrecking natural speech pauses: set it to 0.5 seconds and a brief pause between words (which lasts less) stays intact, and only the longer gaps disappear — the ones that genuinely stretch out the recording without adding anything.
How to adjust depending on the recording
- Interview or podcast with long pauses between question and answer: a moderate threshold (-35 dB) and 0.5-1 second minimum duration, to remove the obvious gaps without touching the conversation's natural rhythm.
- Dictation or voice note with a lot of "uh" and silence while thinking: a lower minimum duration (0.3-0.4s) to be more aggressive.
- Recording with constant background noise (fan, traffic): raise the threshold (closer to 0 dB) because real "silence" never fully drops, or the algorithm won't detect anything.
Frequently asked questions
Does this cut normal speaking pauses? Not if you set the minimum duration well — that's exactly the parameter meant to leave short, natural speech pauses untouched.
Why doesn't it detect any silence in my recording? The threshold is probably too low (too negative) for your recording's background noise level. Raise the value (closer to 0 dB) so it detects more stretches as silence.
How much shorter does the total duration get? It depends on how much real silence the original recording has; the tool shows the duration before and after so you can compare directly.
Is it processed in my browser? Yes, with FFmpeg compiled to WebAssembly, with nothing uploaded to any server.
Shorten your recordings by automatically removing silence with the silence remover, with adjustable threshold and minimum duration, 100% in your browser.