Silence & Gap Remover

Knowledge Base

How Silence Detection Works Here

The tool measures short-window RMS level across your entire file and flags any stretch that stays below your threshold for longer than the minimum duration you set. Rather than deleting those stretches entirely (which can feel jarring), it shortens each one down to your chosen padding length, keeping a natural breath or pause rather than an unnatural jump cut.

Choosing a Threshold

-40 dBFS is a reasonable starting point for most speech recordings with a clean noise floor. Lower it (more negative) if your room tone is quiet and you want to catch only near-total silence. Raise it if there's background noise you also want treated as "silence" between sentences.

Padding Prevents Jarring Cuts

Removing 100% of a pause creates an unnaturally rapid pace that can sound edited and abrupt. Keeping 100-200ms of padding preserves a natural breath and rhythm while still meaningfully tightening a recording full of long pauses.

FAQ

Q: Will this remove breaths and mouth sounds along with silence?
A: Only if they fall below your threshold. Most breath sounds register above typical silence thresholds and will be preserved, which is usually what you want for natural-sounding speech.

Q: Can I undo this if it cuts too aggressively?
A: This tool doesn't modify your original file. Download the result and compare against your source, and simply re-run with different settings if needed.

Setting the threshold and keeping the result natural

Two settings do the work. The threshold decides what counts as silence, measured in decibels below full scale. The minimum duration decides how long a quiet stretch has to last before it is removed. Getting the first wrong removes speech, and getting the second wrong makes the result sound chopped.

Start the threshold just above your noise floor. If the room sits at about minus 50 decibels, minus 40 is a reasonable starting point. Set it too high and the tool eats quiet consonants and the ends of words, because an f or an s is far quieter than a vowel. Set it too low and it finds no silence at all. Check the result on a passage where somebody trails off, since that is where the setting shows.

Padding is what keeps the result sounding human. Removing every pause completely produces speech with no breathing room, which listeners find exhausting within a minute even when they cannot say why. Leaving 100 to 200 milliseconds either side of each cut preserves the rhythm of natural speech while still removing the dead air. Aim to tighten pauses rather than eliminate them.

This works best on single-speaker material recorded in one consistent space. On a conversation with two microphones, removing silence from one track breaks its alignment with the other and the edit falls apart. On music, or anything where a reverb tail runs into the silence, automatic removal cuts the tail off and leaves an abrupt hole where the space used to be.

Vocabulary

dBFS. Decibels relative to full scale, the loudest level a digital file can hold without distorting. Every value is zero or below, so minus 40 dBFS is a quiet sound and minus 6 dBFS is a loud one.

Threshold. The level below which a tool treats sound as silence. Set relative to the noise floor of the specific recording, not to a fixed number, because every room and every microphone sits at a different level.

Padding. A short piece of the original quiet material deliberately kept either side of a cut. Without it, speech runs together unnaturally. Around 100 to 200 milliseconds preserves the rhythm listeners expect.

Dead air. A pause long enough that a listener wonders whether playback has stopped. In spoken word anything past about two seconds reads as a mistake, which is what this tool is for.

Hysteresis. Using a slightly different threshold for entering and leaving silence, so a level hovering near the boundary does not toggle repeatedly. Prevents the chattering that otherwise produces dozens of tiny cuts in one passage.