The three settings, and what each one fixes
Threshold decides what counts as silence. Real recordings are never truly silent — there is room tone, breath, computer fan. Around −45 dB works for a decent recording; a noisy one needs −35 or higher, and setting it too high starts eating quiet speech.
Gap length stops natural pauses being removed. Speech has gaps of a fifth of a second between phrases; cutting those makes the result sound rushed and unnatural. Start at 0.4 s and only go lower if you want it aggressive.
Padding leaves a sliver of silence at each cut. Without it, words butt straight against each other and the edit is audible. A tenth of a second is usually enough to make cuts disappear.
The red bands are the cuts
Every region marked on the waveform is going to be removed. That preview is the point — thresholds are hard to guess and easy to verify. If the bands are landing inside speech, the threshold is too high; if long pauses are unmarked, it is too low.
Detection runs on a mono mix of every channel, so a gap has to be quiet on all of them before it is cut. That prevents an edit that silences one side of a stereo recording.
Where this saves the most time
Podcast and interview recordings, where a two-hour session has twenty minutes of nothing in it. Voice-overs recorded in one take with retries between lines. Lecture recordings with long gaps while the speaker writes.
It is the wrong tool for music: the gaps between notes are the music, and removing them destroys the timing.
Frequently asked questions
Can I preview before downloading?
Yes — the preview button plays the trimmed result straight away, so you can hear whether the cuts sound natural before spending time on the export.
Is my audio uploaded?
No. The analysis and the edit both happen in the page — the file never leaves your device.
Something wrong with this tool, or an idea for it? Tell us