Noise Reduction
Clean up a recording. Voice mode removes background noise around speech; steady-noise mode learns a noise sample and subtracts it.
MP3, WAV, M4A, FLAC, OGG, Opus, AIFF, WMA, AMR and the sound of most videos. Files stay on your device.
How it works
- Choose a recording.
- Pick Voice (speech) or Steady noise, and set the strength.
- Preview and compare with the original.
- Export and download.
Two ways to remove noise
Voice mode runs RNNoise, a small neural network trained on speech, entirely on your device. It keeps what sounds like a voice and turns down almost everything else: fans, traffic, keyboards, air conditioning, room tone. A strength slider blends the original back in if the result sounds too processed.
Steady noise mode is spectral gating. It learns what the noise looks like (its level at every frequency), then turns down any part of the sound that matches it. It works for constant noise such as tape hiss, a hum floor or a computer fan, and it works on music as well as speech. It does not help with changing sounds like passing cars or other people talking.
Teaching it the noise
In steady noise mode you can let the tool find the noise itself, or select a part of the waveform that has only the noise in it and press the button. Automatic mode takes the quietest half-second stretches of the file, so it needs a few moments with only the noise (a pause before speaking is enough). A selected sample must be at least 0.2 seconds; one or two seconds gives a steadier result.
Noise reduction sets how far the noise is turned down, sensitivity how far above the noise a sound must rise to be kept, and smoothing hides the glitchy "musical noise" that gating can leave. Keeping a small share of the noise often sounds more natural than a perfectly silent background. Switch the result to "Only what is removed" to hear exactly what is being taken away; if you hear parts of the voice or music in it, lower the reduction or the sensitivity.
What it cannot do
Noise reduction trades noise for artifacts, and the more you remove the more of them you get. It cannot restore speech that is buried in noise, clipped or distorted. Voice mode is made for speech: music, singing and other sounds are treated as noise and can be damaged, so do not use it on a song. Stereo files are cleaned one channel at a time. Always compare with the original using Preview before you export.
Questions
Which mode should I choose for a podcast or voice memo?
Start with Voice. It handles most background noise around speech without a noise sample. Use Steady noise if the problem is a constant hiss or hum, or if the recording is not speech.
Why does the voice sound thin or watery afterwards?
The reduction is too strong. In Voice mode lower the strength; in Steady noise mode lower the reduction or sensitivity, raise smoothing, or keep a little of the noise.
Can it remove music or other voices from the background?
No. Voice mode removes sounds that do not look like a single voice, but background speech and music are often left or partly kept. For a song, use the vocal remover instead.
Does it change the length of my file?
No. The result has exactly the same length as the original and lines up with it sample for sample.
Does it work on long recordings?
Yes. The file is processed in pieces, so memory stays small. The speech model (about 5 MB) loads the first time you use Voice mode.
Is my audio uploaded?
No. Everything runs in your browser and the file never leaves your device.