Drop an audio file with speech — a voice memo, podcast or interview.
Full strength removes the most noise. Lower it to keep some natural ambience.
Listen to the result and download an MP3 or WAV.
RNNoise is a small neural network made for voice. It is loaded only when you use it.
Recordings are processed in your browser.
Blend the cleaned and original sound to avoid a processed feel.
Export a small MP3 or a lossless WAV at 48 kHz.
Noise reduction here uses RNNoise, a small neural network trained to separate speech from background sound. It works well on hiss, fan and air-conditioner hum, keyboard clicks and general room or street noise behind a voice.
It is made for speech. Music, singing and ambient recordings can sound thin or damaged, so use another approach for those. It will not remove other people talking in the background, and it cannot fix a recording that was clipped or distorted.
At 100% you get the strongest reduction. If the voice sounds a little processed or “underwater”, lower the strength to blend some of the original back in. Stopping at 70–85% often sounds more natural.
Record close to the microphone, in a quiet room, with soft surfaces around you. Software can only reduce noise; a good recording starts with less noise. Normalize the result afterwards if the level dropped.
The cleaned audio is processed at 48 kHz. Stereo files are handled channel by channel. Export WAV if you are going to edit further, or MP3 for sharing.
The model is about 120 KB and is loaded only when you press the button. Your file is not uploaded to any server.
Resize, crop, convert and clean up images, audio and video — all in your browser.
Browse all tools →