Introduction

NVIDIA recently released RTX Voice, a Windows application—or plugin, as they call it—which taps into the AI capabilities of Tensor Cores. Tensor Cores were originally one of the key selling points of the Turing GPU microarchitecture (GeForce RTX 20xx family) launched in 2018. They date back even further as they were first implemented in the Volta microarchitecture (Titan V and Quadro GV100). If you have ever bothered researching their purpose, you probably learned that they handle complex 4x4 matrices and are designed to have superior deep learning neural-net building and training performance that is up to 12 times higher than with "regular" CUDA cores. NVIDIA's technical documents relating to Tensor Cores mention several other buzzwords, such as machine learning and artificial intelligence, but chances are the majority of GeForce RTX users still haven't got a clue about the exact role of Tensor Cores in their expensive graphics cards, or whether they are good for anything at all. With RTX Voice (currently in beta), Tensor Cores are finally in charge of something tangible and potentially useful to just about anyone. This suddenly puts the RTX technology on the table not just for gamers, but for businesses and enterprises as well.



The main idea behind RTX Voice is to use artificial intelligence to remove background noise from any existing audio input (microphone) and/or output (speakers or headphones). NVIDIA's timing couldn't be better. Due to the COVID-19 pandemic and quarantines, the number of users of video conferencing and VoIP apps is rapidly growing. Just as importantly, more and more gamers are looking to launch their Twitch/YouTube streaming careers. All of those could benefit from a technology that makes sure their audience doesn't hear the click-clacking of their mechanical keyboard, whirring of system fans, voices of nearby family members, as well as pretty much any other type of background noise that could be picked up by the microphone listener(s) might find annoying. In this mini review, we provide a practical test of RTX Voice with a selection of different kinds of microphones and headsets in various real-world background noise scenarios to figure out the impact RTX Voice can have.

Once again, RTX Voice is currently in beta. NVIDIA invites everyone with an RTX graphics card to try it out and give their feedback. Participating in the beta doesn't require you to register. You're not obliged to share your thoughts or data either and can try it out in action even if you don't own a microphone, so if you're interested in the technology, visit the official website and follow the detailed setup guide.

NVIDIA will gladly accept your help to train their machine learning AI network for RTX Voice. If you're willing to assist, you can visit the RTX Broadcast Engine Feedback website, record yourself speaking for 15 seconds (there's a supplied piece of text), record your surroundings, and send them the samples.

System Requirements

To use RTX Voice, you'll need a NVIDIA GeForce RTX or Quadro RTX graphics card, the latest version of the official video driver (410.18 or newer), and an up-to-date installation of Windows 10. You'll also have to do a bit of tweaking in the video conferencing and VoIP apps you plan to use. You'll have to make sure you turn off any built-in microphone noise canceling effects as they could clash with what RTX Voice is trying to achieve, resulting in a lower quality end result. For example, in Discord you'll have to visit the settings, go to "Voice & Video", scroll down to "Advanced" and uncheck "Noise Suppression", "Echo Cancellation", "Noise Reduction", and "Automatic Gain Control". Also check your sound card drivers and turn off any microphone filtering effects there if they exist.

Supported Apps

• OBS Studio
• XSplit Broadcaster
• XSplit Gamecaster
• Twitch Studio
• Discord
• Google Chrome
• Battle.net Chat
• WebEx*
• Skype*
• Zoom*
• Slack*
• Steam Chat**

* RTX Voice speaker output may exhibit issues in these apps
** Turn off "Noise Cancellation" for better quality