Built on Xiph.Org's deep learning framework , it gives creators, gamers, and remote professionals real-time, system-wide background noise suppression without relying on expensive, proprietary hardware or heavy software suites.
| OS / Platform | Library Name | |---------------|--------------| | Windows 32-bit | librnnoise-windows-x86.dll | | Windows 64-bit | librnnoise-windows-x86-64.dll (commonly named as above) | | Linux | librnnoise-linux-x86-64.so | | macOS ARM | librnnoise-macos-aarch64.dylib | | macOS Intel | librnnoise-macos-x86-64.dylib |
You might be thinking, "My DAW already has a noise gate," or "Zoom has a noise suppression button." Why bother with a DLL file? librnnoisevstdll
Copy the .dll file into that folder.
NoiseTorch-ng uses RNNoise to create a virtual filtered microphone that any application can use: Built on Xiph
RNNoise—originally designed by Jean-Marc Valin of the Xiph.Org Foundation—takes a completely different approach. It utilizes a specifically trained on thousands of hours of clean speech paired with real-world noise profiles. The algorithm learns the defining characteristics of human speech, allowing it to mathematically separate and strip away non-voice audio artifacts in real-time.
Assumptions (reasonable defaults):
Set your system audio and OBS audio settings to 48kHz (48000 Hz).
is an open-source noise suppression library based on Deep Learning (recurrent neural networks). It was developed by Jean-Marc Valin (of Mozilla’s Daala and Opus codec fame). Unlike traditional noise gates, which simply cut audio below a certain volume threshold, RNNoise is "smart." It has been trained on thousands of hours of audio to recognize the difference between human speech and background noise. NoiseTorch-ng uses RNNoise to create a virtual filtered
Below is a complete implementation example showing how to integrate librnnoisevstdll into a Windows C++ application. This example assumes the DLL is present in your executable directory.
: A noise suppression algorithm created by Jean-Marc Valin of the Xiph.Org Foundation. Unlike classic subtractive filters that target specific static frequencies (like tape hiss), RNNoise uses a Recurrent Neural Network (RNN) . This deep-learning architecture is trained specifically to distinguish complex human speech from unpredictable background noise.