The futuristic Turing GPUs inside of Nvidia’s GeForce RTX 20-series graphics cards come packed with dedicated RT and tensor cores devoted to accelerating ray tracing and AI tasks, respectively. Thus far, the tensor cores have mostly been used to supercharge visuals by reducing noise in ray traced games and enabling the faster, better, deeply impressive DLSS 2.0 feature. But a new Nvidia app called RTX Voice puts those machine learning chops to a much more practical use for your ears, leveraging the hardware to filter out background noise in conferencing apps like Zoom and Skype, as well as broadcasting apps like OBS Studio, Discord, and Xsplit.
It’s a perfect release for the current time, where so many of us are struggling with a new work-from-home normal, rife with inferior webcams and loud surroundings, while many more are putting their spare time to good use by dabbling in content creation.
Background-noise filtering is nothing new. Plenty of apps offer to do it, albeit with notoriously mixed results. By shifting the workload away from software and onto the tensor core hardware, Nvidia’s solution is much more potent, even in its current beta form.
“Nvidia gave creators superpowers, basically,” said EposVox, a media creation and streaming expert, in his April 16 video detailing RTX Voice and other Nvidia technologies. I’ve played around with the technology a bit and find myself just as impressed. It did a magical job at eliminating the sound of my noisy Cherry MX Blue keyboard during Skype and Discord calls.
How to use Nvidia’s RTX Voice
You’ll need an up-to-date, fairly modern system to be able to take advantage of RTX Voice. GeForce RTX 20-series graphics cards or their Quadro RTX equivalents are an obvious requirement, as they’re the only Nvidia hardware that includes the crucial tensor cores right now.
You’ll also need Windows 10, the RTX Voice beta app (warning: direct download link), and current drivers for your video card. The company’s explainer lists Nvidia Studio drivers version 410.18 as the standard, but if you don’t run the creator-focused Studio drivers, it also worked with the newest GeForce Game Ready drivers (version 445.87) on my own system. Just make sure you’re running the latest drivers for whatever flavor you use, basically.
Brad Chacos/IDG
The RTX Voice beta’s setup is simple and straightforward. (Click to enlarge.)
Once you’ve installed the RTX Voice software, you need launch it to configure it. It’s pretty straightforward. Tell it your primary input (mic) and output (speakers/headset) sources, select whether to enable noise filtering for each, and use the slider bars to adjust how aggressively the noise filter behaves.
These are key decisions, though you can always open the app again to tweak the settings. Because RTX Voice uses your graphics card to process audio, it can have a performance impact, especially if you’re simultaneously gaming while using an application that taps into RTX Voice. We haven’t had an opportunity to benchmark the impact ourselves, but Epos Vox says that in his testing, having RTX Voice active on just your input mic can reduce your gaming frame rate by 4 to 10 percent, or all the way up to about 20 percent if you have both the mic and speaker performing noise cancellation.
That shouldn’t matter if you plan to use RTX Voice for Zoom or Skype calls only while you’re working. But it may matter if you plan to use the technology to clean up your inbound audio while you’re running Discord or OBS, though the degree of impact will obviously vary game-to-game and system-to-system.
If you’re only worried about your own mic sounds, disable RTX Voice’s noise filtering on your audio output to keep the performance hit small. You only want to enable RTX Voice for your headset or speaker if you want to actively filter out background noise coming in from other people’s microphones in Zoom, Discord, or whatever. That could be handy in worktime conference calls or if you’re trying to capture clean audio for a gaming video, but Nvidia suggests leaving it disabled in most cases.
Brad Chacos/IDG
RTX Voice utilizes the tensor cores exclusive to GeForce and Quadro RTX graphics cards, like the GeForce RTX 2060.
“We recommend turning on RTX Voice for your microphone, and turn it on for your speakers only if needed,” the setup guide says. “While quality remains almost the same, it is possible to see a small difference in some cases, and the solution takes system resources that you can avoid using if you don’t need it.”
Got it? Good. RTX Voice creates a virtual audio device on your system, which will now run constantly in the background unless you manually disable it in Windows 10’s system tray.
You’re not done, however. To use RTX Voice, you’ll need to set up each individual piece of supported software to use it. The RTX Voice beta currently supports 10 popular programs:
- OBS Studio
- XSplit Broadcaster
- XSplit Gamecaster
- Twitch Studio
- Discord
- Google Chrome
- WebEx*
- Skype*
- Zoom*
- Slack*
To enable RTX Voice, you’ll need to go into the settings for each app and configure their input and output away from your defaults, pointing them to the new “Nvidia RTX Voice” option instead. If you need help finding the settings, Nvidia’s setup guide can walk you through the steps for each individual application, and also provides detailed information on how you can test the efficiency of the input and output filtering independently on your own system. Note that you’ll usually want to disable any software-based noise filtering the individual programs offer if you plan on using RTX Voice.
Brad Chacos/IDG
You can see “Nvidia RTX Voice” as a selectable microphone and speaker option in Discord here.
Entries marked with an asterisk might run into issues with output (speaker) audio filtering during the beta, Nvidia says—a bit of a bummer, because they’re the videoconferencing apps probably most susceptible to have kids screaming or morons clacking on keyboards in the background. Alas. Most people should probably leave output filtering disabled anyway.
When you enable noise filtering on your own mic input, your own screaming children, clattering keyboard, and taco farts will be eradicated with startling accuracy, at least in my own quick Skype and Discord tests. Watch the Epos Vox video embedded at the top of this article to see how well RTX Voice filters out the sound of a handheld vacuum he waved around his microphone—it’s eye-opening.
A few final notes: The RTX Voice beta doesn’t filter your Windows system noise, only audio captured via your input device, so your game sounds and Spotify tunes (for example) won’t be affected. The beta also doesn’t manage the volume of your mic and speaker. “If you are testing RTX Voice (Speakers) and want to adjust the volume, go back to your previous device, adjust the volume, and go back to RTX Voice,” Nvidia says. Lastly, remember that this is still in beta, so you may see bugs or other unanticipated behavior.
Nvidia
The tensor cores in Nvidia’s Turing GPUs aren’t just for ray tracing anymore.
That said, RTX Voice has worked like a charm in the limited testing I’ve performed with it in conferencing apps this afternoon. It’s incredibly impressive even in beta form. If you’ve got a GeForce or Quadro RTX graphics card and use any of the supported apps, why not give it a shot?
And if you’re a content creator, this isn’t the only way that Nvidia’s leveraging tensor cores to make your life easier. Check out our coverage of Nvidia’s RTX Broadcast Engine technology to see how GeForce’s AI smarts can get you real-time virtual greenscreens, augmented reality effects, and style filters in supported software.