Voice
Denoiser

The tool runs RNNoise, a recurrent neural network trained to separate speech from background noise, compiled to WebAssembly. It analyzes each 10 ms frame, estimates how much of the signal is speech versus noise, and suppresses the noise component. Because the network runs locally in your browser, a 5-minute recording processes in seconds — with nothing uploaded.

No. Recording and processing happen entirely in your browser — the audio never leaves your device. There is no server, no account, no tracking, and no data collection. This matters for sensitive material like client calls, unreleased content, or medical recordings.

RNNoise is trained for speech, so it's the right tool for voice recordings, podcasts, voiceovers, and video dialogue. It is not designed for music — a trained speech model will treat music as noise and damage it. Use it on anything with a human voice and leave music alone.

Whatever your browser can decode via the Web Audio API — typically MP3, WAV, M4A/AAC, OGG Vorbis, and FLAC. The audio is downmixed to mono for processing (RNNoise is a single-channel speech model) and exported as a 48 kHz WAV file.

In controlled tests, RNNoise typically removes 10-15 dB of background noise while keeping speech intelligible — enough to clean up fans, traffic, room hum, and keyboard sounds. It will not remove loud music, overlapping talkers, or extremely heavy distortion. For extreme cases, run the file through twice or combine with a noise gate.

Yes — completely free, with no account, no watermark, and no usage limits. RNNoise itself is open source under a BSD-style license, and this tool is a privacy-first wrapper around it. Alternatives like Adobe Podcast require an account and upload, and Krisp is a paid subscription.