One click runs a small on-device voice detector (about 0.1 MB, size shown before the one-time download, cached for next time) that keeps every spoken part and removes the gaps between them.
Remove Silence from Audio - Auto Dead-Air Cutter, On-Device
Cut the silent gaps - long pauses, dead air, gaps between takes - out of a voice recording automatically, without the audio ever leaving your browser. A small on-device voice detector listens through the file, keeps every spoken part, and removes the gaps between them.
Key features
- A small on-device voice detector runs the cut - about 0.1 MB, its size shown before the one-time download, and cached for your next visit.
- You choose how long a pause has to be before it counts as dead air - 0.3, 0.5, or 1 second - then press one button.
- The kept audio is your original recording, untouched: the detector only chooses where to cut, it never changes the sound in between.
- The result appears as two players side by side - the original and the trimmed version - with a plain summary of how many seconds of dead air were removed across how many gaps.
- The output is always a mono 48 kHz WAV file, whatever the input format was.
When to use it
A raw podcast take where you paused to check notes; a lecture or meeting recording with long stretches of nobody talking; a voice memo you started and then left running. Cutting that dead air by hand in a waveform editor means finding every gap yourself - here you pick the file, choose the minimum pause, and press one button. Listen to the trimmed version first, then download. If the cut feels too tight, pick a longer minimum pause and run it again; each pass starts from your original file.
How the cut is decided
The detector is a small neural voice-activity model. It scores every 10 ms of audio for the probability that someone is speaking; stretches below the line that run longer than your chosen minimum are removed. Speech is always kept with 0.15 s of breathing room on each side, and pauses shorter than your minimum are left exactly where they were, so the result still sounds like a person talking rather than a chopped edit.
Voice detector vs Basic cut
If you would rather not download the engine at all, the Basic cut button does the same job by loudness alone. The button labels say which is which, so you can choose.
| Cut mode | How it judges a frame | Trade-off |
|---|---|---|
| Voice detector | Neural voice-activity probability, scored per 10 ms of audio | One-time ~0.1 MB download; keeps breathy or very quiet speech the loudness test would drop |
| Basic cut | Loudness only - frames quieter than the recording's own noise floor count as silence | Instant and needs zero network, but it can clip breathy or very quiet speech |
Know its limits
The detector is tuned for speech. On music it will treat quiet instrumental passages as silence and cut them - for music editing, use the manual Audio Trimmer, where you pick the cut points yourself. And if your recording also has constant hiss or hum behind the voice, run it through the Audio Denoiser - the same on-device engine family, pointed at noise instead of pauses - before or after cutting.
Privacy
There is no upload, no account, and no server: the recording stays on your device from the moment you pick it to the moment you download the result.
Frequently Asked Questions
Is my recording uploaded anywhere?
No. The file is decoded and processed entirely inside your browser tab. The only network request is a one-time ~0.1 MB download of the voice-detection engine itself - your audio never leaves your device, on either the AI or the Basic path.
Will it cut a word in half or make the speech sound choppy?
It is designed not to: every kept speech segment gets 0.15 s of breathing room on each side, and pauses shorter than your chosen minimum (0.3, 0.5, or 1 second) are left exactly where they were. If a cut still feels too tight, pick a longer minimum and run it again - each pass starts from your original file.
Does it change how my voice sounds?
No. The detector only decides where to cut - the audio it keeps is your original recording, sample for sample. If you also want hiss or hum removed, that is a different job: use the Audio Denoiser page, which runs the same engine family in noise-removal mode.
Can I use it on music or a song?
It will run, but the detector is tuned for speech: quiet instrumental passages score as silence and get cut. For music, use the manual Audio Trimmer instead, where you choose the cut points yourself.
What format does the result download in?
Always a mono 48 kHz WAV file, whatever the input was (MP3, WAV, M4A, OGG, OPUS, WebM, FLAC, or AAC in - WAV out). WAV keeps the trimmed audio lossless; convert it afterwards if you need a smaller file.