Neural network noise filter

In this post I will show you how to improve your microphone audio in video calls with the help of a filter, based on artificial intelligence, which tries to filter everything that is not a voice from your audio. The filter can achieve an effect very similar to NVidia's RTX voice. Hear for yourself what it can do:

Unlike RTX Voice, you don't need to have an Nvidia GPU to use the filter, in fact you don't need a dedicated graphics card at all, as all processing is done by the processor, and the best thing, works very well even on the most modest computers.

Here is the video tutorial:

Downloads

Three pieces of software are necessary to implement this filter, they are detailed below.

The heart of the filter is the “noise-suppression-for-voice” plugin, which is based on the RNNoise library, for noise suppression using recurrent neural networks, it is available at this GitHub project page.

What we are interested in in this project is a VST plugin, VST is a type of plugin widely used in audio processing and, to use the plugin, we need a program capable of running VST plugins.

That's where the second download, VSTHost, comes in. If you are familiar with VST plugins, you can use other software that you are more used to, this was the simplest one I found.

The last program purpose is to transform the audio output of the VSTHost into an input that can be selected in your videoconferencing or recording software, a kind of virtual microphone. The program is called VB-CABLE Virtual Audio Device.

Installation and Configuration

We start by installing the VB-CABLE, in 64 bit systems you need to run the setup file as an administrator, then click on install driver. A message will appear that you may need to restart the computer, not really necessary.

Now open the VSTHost and, to add the filter plugin, we go to the menu “file > new plugin” and navigate to the plugin location, inside the folder where it was unzipped "bin > vst > librnnoise_vst.dll".

The next step is to configure the audio input and output of the VSTHost, for that we have to go to the menu "Devices > wave". In the input port you select the microphone you normally use, which starts with "MME:", for output port have to select "MME: Cable input" or something similar, sometimes variations can occur, look for "vb-cable" in parentheses.

All set, now just have to select the "Cable output" as the audio input in your preferred videoconference software.

If you want to do the inverse, filter the audio of other people in a videoconference, you need to reverse the configuration. Configure the audio output in the video conference software as "Cable input" and, in the VSTHost, select the "Cable output" as the input port and your computer's speakers as the output port.

Unfortunately, with this configuration you cannot do both at the same time, but it is possible, for that you will need an additional virtual audio device, like VB-CABLE A+B, to install it you just need to make a small donation to the developers.

Closing thoughts

In times of semi-mandatory home office, where you end up making several video calls in a single day, this filter can be very useful, blocking diverse sounds like keystrokes, cars passing by, banging on the table, among others. Enjoy =)