Clean Up a PC Mic With Nvidia Broadcast—Without Sacrificing Game FPS

If your gaming headset or inexpensive USB microphone sends keyboard clatter, fan noise, and room echo into every call, an RTX graphics card may already include the upgrade you need. Nvidia Broadcast can turn a physical microphone into a filtered virtual input for Discord, Zoom, Teams, OBS, Google Meet, and other Windows apps. The useful part is not merely switching it on. It is choosing the lightest effect that solves the problem, then making sure another app is not processing the same audio a second time.

Nvidia Broadcast 2.3 is a free Windows application. Nvidia lists a GeForce RTX 2060, Quadro RTX 3000, Titan RTX, RTX Spark, or newer GPU as the baseline, along with 8GB of system memory and driver version 572.61 or later. Windows 10 and Windows 11 are supported, and installation needs an internet connection because the app downloads the AI models required by the installed GPU.

That hardware requirement is worth checking before you download it. The Broadcast app does not officially run on GeForce GTX cards. Nvidia still offers its older RTX Voice noise-removal path for some GTX hardware, but the company warns that results vary. Studio Voice, one of Broadcast's heavier microphone effects, raises the floor to a GeForce RTX 3060 or better.

Build the audio chain in the right order

Start in Nvidia Broadcast, open the Microphone tab, and choose your actual headset or microphone as the source. Enable Noise Removal, speak into the microphone, and use the built-in recording sample to compare the processed voice with the original. Add Room Echo Removal only if the room itself sounds hollow or reflective. Noise Removal targets sounds around you; Room Echo Removal targets reverberation from your voice bouncing around the space.

Next, open the voice or streaming app and select Microphone (Nvidia Broadcast) as its input. Broadcast creates this virtual device so the communication app receives the processed signal rather than the raw microphone feed. If people still hear the unfiltered mic, the most common cause is that Discord, OBS, Zoom, or another program is still listening directly to the physical device.

Do not choose the Nvidia Broadcast virtual microphone as the source inside Nvidia Broadcast itself. That creates a loop. The source in Broadcast should be the real device; the input in the destination app should be the virtual Broadcast device.

There is also little value in stacking two noise-removal systems. Nvidia specifically recommends disabling similar processing in the communication app or device driver. If Discord's noise suppression, a headset vendor's filter, and Broadcast all work on the same voice, the result can sound thin, clipped, or unstable. Keep one processor in charge, then test again with a recording rather than judging your own voice through live monitoring.

Noise Removal is the gaming setting; Studio Voice is not

Studio Voice applies a more intensive AI equalizer and offers Default, Bright, Full, and Warm microphone profiles. It can give a spoken presentation or recorded segment a more produced tone, but Nvidia describes it as a feature for video conferencing, content creation, and non-gaming livestreams. The company does not recommend using it while gaming or running another GPU-heavy application because it consumes a larger share of GPU resources.

That distinction matters on a gaming PC. A GPU near full utilization has little spare capacity for extra camera and voice effects. Nvidia says Broadcast's performance cost depends on the GPU, game, webcam resolution, frame rate, and number of active effects. Its published 5% to 10% example was measured on an RTX 5090, so owners of lower-tier cards should not treat that figure as a universal ceiling.

For an in-game voice chat, begin with microphone Noise Removal alone. Leave Studio Voice, webcam background replacement, eye contact, and other video effects off unless you actually need them. If a stream still drops frames, cap the game's frame rate or lower its GPU load, then retest. Reducing a camera to 720p at 30 frames per second can also help when video effects are necessary.

Broadcast 2.3 allows multiple effects, but support for combining them does not make every combination wise. Nvidia notes that more effects increase system load, and some effects cannot run together. Studio Voice, for example, cannot be combined with microphone background-noise removal. Pick the problem you need to solve rather than turning on every switch.

Know what software cannot repair

Noise processing works best when it receives a usable voice signal. It cannot restore words that were clipped because microphone gain was too high, nor can it create detail that a microphone never captured. Move the microphone closer to your mouth, aim it correctly, lower the input level if loud speech distorts, and keep it away from a direct blast of PC or air-conditioner airflow. Those changes improve the signal before the AI has to separate voice from noise.

Room Echo Removal also has limits. Heavy echo may call for a softer room, curtains, a rug, or a different microphone position. Combining echo removal and noise removal can sometimes reduce cancellation quality; Nvidia's troubleshooting guidance says the two effects work together but acknowledges that the beta echo filter may weaken noise removal in some cases.

Finally, configure Windows carefully if you want Broadcast to work everywhere. Setting the virtual Broadcast microphone as the Windows default input is supported, provided Nvidia Broadcast itself points to the real microphone. Do not set the virtual Broadcast speakers as the default output, because that can filter general system audio such as music and video. Per-app selection is less convenient, but it makes the signal path easier to diagnose.

The sensible setup is small: one physical microphone, one Broadcast effect, and one virtual input selected in the app that needs it. That can remove the noises people complain about without turning a voice call into another demanding graphics workload.