RTX Voice Beta Vows to Suppress Audio Background Noise Through AI (Tensor Cores)


RTX Voice is a new plugin (downloadable here) currently in beta testing that aims to exploit the AI power (presumably through Tensor Cores) of NVIDIA's GeForce and Quadro RTX graphics cards to suppress any audio background noise.

How does it work? RTX Voice creates a virtual audio device on your Windows 10 PC. After installing the app, you'll have to point it to your microphone and/or speakers (yes, it can also remove background noise from incoming audio, not just outgoing audio). However, it is important to point out that you should not be using the RTX Voice virtual audio device as the Windows 10 default on your PC. That's because doing so would also filter audio sources that don't really need to be denoised, such as game audio or music; you'll have to select the RTX Voice virtual audio device only for your voice chat applications.

NVIDIA GeForce RTX 4080 Graphics Card Specs, Performance, Price & Availability – Everything We Know So Far

The beta already supports some of the most popular voice chatting and broadcasting applications, such as:

  • OBS Studio
  • XSplit Broadcaster
  • XSplit Gamecaster
  • Twitch Studio
  • Discord
  • Google Chrome
  • WebEx
  • Skype
  • Zoom
  • Slack

The last four on that list, though, may have issues while suppressing speaker output. Indeed, NVIDIA's setup guide mentions that you should probably only activate speaker output denoising when really needed, as there could be a slight audio quality difference in some instances and the RTX Voice app would needlessly take some resources from your system.

It is also important to ensure your software's built-in noise cancelling effects are turned off while you have RTX Voice enabled. Do note that you won't be able to control the volume of your underlying devices, which means to adjust the volume you'll have to select your previous devices, tweak it, and then select RTX Voice once again.

You can also help NVIDIA train their AI network by recording yourself speaking for 15 seconds, then record the surroundings, and submit the audio here.