Pick a voice recording and get it back with the fan, traffic or hiss behind it turned down. A video comes back as a video, with the clean sound in place.
Drop a recording here
Audio or video, up to 10 minutes. Private: your files stay on your device.
No recording yet? Record one in the mic test
It's built for speech: voice messages, calls, lectures, podcasts and voice-overs.
Standard is RNNoise, a small neural network from Xiph.org, the makers of the Opus codec. Every 10 milliseconds it splits the sound into frequency bands, estimates how much of each band is voice, and turns down the rest. Stronger is DeepFilterNet3, a larger network that removes more noise and keeps more of the voice when the noise is loud.
Steady noise behind a voice: computer and air conditioner fans, traffic, rain, the hiss of a cheap mic or preamp, the drone of a fridge or a car.
Other people talking, music, echo from a bare room, distortion from a level set too high, and loud sudden sounds like a dog's bark. Music in the file gets damaged too, because it isn't speech.
The filter downloads once, about 3 MB, or 11 MB more if you choose the stronger one, and runs in your browser. Your recording stays on your device, and the cleaned file is made there too.