How to use
- Add a recording.
- Choose Spectral gate (steady hiss and hum) or Voice (RNNoise, for speech).
- Optionally select a noise-only part and use it as the profile; set the strength, compare A/B, and export.
Worked example
On the sample (a voice over fan noise), Spectral gate at 12 dB reduction lowers the noise floor by about 10 dB while the voice stays clear.
Supported formats and limits
| Input | MP3, WAV, M4A, Ogg, FLAC, WebM |
|---|---|
| Output | WAV, FLAC, MP3, M4A (AAC), Ogg Opus |
| Limits | About 23 minutes of 48 kHz stereo (46 minutes mono); decoded into memory. |
| Engine | STFT spectral gating (noise profile from a selected region or the quietest frames) and RNNoise (WebAssembly) in a Web Worker |
Limitations
- Strong settings cause a watery “musical noise” artifact.
- Sudden noises (door slams, keyboard clicks) are not removed reliably.
Questions
How do I get a cleaner result?
Select a second or more where only the noise is audible; that becomes the noise profile. Without a selection, the quietest 10% of the recording is used. Around 10 to 15 dB of reduction sounds natural; stronger settings can cause a watery, chirpy sound.
Which method should I use?
Spectral gate suits steady noise like fans, hum or hiss and uses the noise profile. Voice (RNNoise) is a neural suppressor trained for speech; it works at 48 kHz, so its result is 48 kHz. Neither removes sudden noises such as door slams or keyboard clicks reliably.
Privacy
Runs on your device. Files and text are processed in this browser tab and are not uploaded.
See the privacy policy for how toolsdocks handles data.