RNNoise 0.2 Released With AVX2 Optimizations For Neural Network Noise Suppression

Xiph.Org has released RNNoise 0.2 as the recurrent neural network for audio noise reduction. This library leverages a neural network model for enhancing real-time noise suppression.

RNNoise 0.2 was released today as the latest iteration of this deep learning powered real-time noise suppression library. There are many improvements along with improved training for it. Most notable for end-users are likely the SSE4.1 and AVX2 optimizations for better performance. There is also RTCD-powered CPU runtime detection for determining the optimal features to use at run-time.

RNNoise 0.2 has also been changed so that the distributed models are trained using only publicly-available datasets.

Downloads and more information on the RNNoise 0.2 release via GitHub. Those learning about this RNNoise effort for the first time that is supported by Mozilla and Xiph.Org can learn more background information here.

I'll be working to get out some updated RNNoise benchmarks of different processors soon that should be interesting given the AVX2 optimizations in this v0.2 release.