Why fixed silence levels cut words
Every silence remover asks two things: how quiet counts as silence, and how long it has to last. Most use a fixed level, such as −20 dB (the default in Audacity’s Truncate Silence) or the −30 and −50 dB levels in the FFmpeg commands that are copied around forums. A fixed level does not know how loud your recording is, so on a quiet recording, or a speaker who drops their voice, it treats speech as silence.
We tested this on a 54-second voice recording: eight sentences (39.3 seconds of speech)
with pauses of 0.35 to 4 seconds between them, background noise at −58 dB, and the fifth
sentence spoken 22 dB quieter, like an aside. We ran FFmpeg 7.1’s
silenceremove filter with three common settings and this page with its
defaults, then searched each result for every sentence.
| Setting | Result length | Quiet sentence | Normal sentences |
|---|---|---|---|
| FFmpeg, −20 dB, pauses over 0.5 s | 36.8 s | Gone | Chopped between words |
| FFmpeg, −30 dB, pauses over 0.5 s | 37.0 s | Gone | Chopped between words |
| FFmpeg, −50 dB, pauses over 1 s | 47.2 s | Kept | Kept, with up to a second of each pause left |
| This page, automatic level, pauses over 0.5 s to 0.3 s | 41.4 s | Kept | Kept |
The first two results are shorter than the speech itself: they removed 2.3 to 2.6 seconds of words. The automatic level here measured this recording’s background noise in 10 ms steps (−61 dB) and set silence at −47 dB, below the quiet sentence and above the noise. You can still set the level yourself with the slider; the dashed line on the waveform shows where it sits.
The three settings
| Setting | Default | Change it when |
|---|---|---|
| Shorten pauses longer than | 0.5 s | Raise it to 1–2 s to remove only dead air and keep natural pauses between sentences; lower it for a fast, jump-cut pace. |
| Leave a pause of | 0.3 s | 0.1 s gives a tight, jump-cut pace; 0.4–0.6 s sounds unhurried. It does not go to zero: soft word starts and endings sit below the silence level, and in our test joining the sound end to end clipped them in the quiet sentence. |
| Silence is quieter than | auto | Move it right if breathing, fan noise or room hum is being kept; left if soft words turn red on the waveform. |
Pauses are cut in the middle, so a short part of each pause stays on both sides: the breath after a sentence and the breath before the next one. The start and end of the recording are trimmed to a tenth of a second unless you untick that option.
What the check proves
After saving, the page decodes the new file and compares it with the plan and with the original. The length has to match the plan to the sample, and every kept part has to match the original sound. In a WAV file everything outside the 10 ms blends at the joins is identical, sample for sample. MP3 and M4A are compressed again, so the page reports how far below the sound the encoding difference is instead.
Podcasts, lectures and video
For a podcast or interview, the defaults remove dead air without changing how people talk. For a lecture or a voice memo you will listen to at speed, combine this with Change audio speed: remove the pauses first, then speed up the talking without changing the pitch. To edit a video the same way, download the cut list and apply the cuts in your editor; the times are in the original recording’s timeline.
Related tools
To cut a recording at its pauses into separate files instead, use Split audio. To cut off one part at the start or end, Trim audio. If the speech is too quiet overall, Increase audio volume raises it without clipping, and Voice recorder records with the browser’s noise filters off so quiet sounds are not faded out in the first place.