Uses deep learning trained on community-submitted noise samples to isolate voices dynamically.
RNNoise is a noise suppression library developed by Jean-Marc Valin and the Xiph.Org Foundation. Unlike traditional noise gates that simply mute audio below a certain volume threshold, RNNoise uses a recurrent neural network (RNN). It is trained on thousands of hours of noise and clean speech to instantly distinguish between your voice and unwanted background sounds. The VST Implementation librnnoise-vst.dll
The .dll version of this library allows it to function as a plugin. This standard format allows it to be integrated into various Digital Audio Workstations (DAWs) and streaming software: It is trained on thousands of hours of
To appreciate the file, you must appreciate the algorithm. Traditional noise suppression (like ReaFir or Audacity’s Noise Reduction) requires a "noise print"—a sample of pure background noise. RNNoise does not need that. you must appreciate the algorithm.
Aggressive antivirus (particularly Windows Defender or Malwarebytes) sometimes flags unknown DLLs as "Potentially Unwanted Programs" because they hook deep into audio processes.
Unlike standard filters that use static algorithms, RNNoise uses a Gated Recurrent Unit (GRU) to "learn" the difference between speech and noise.
Download the latest release for Windows. Extract the archive and locate librnnoise_vst.dll — this is the file you will use.