Student project · public preview v0.2.0
RTX voice processing on Linux
I made this because I wanted NVIDIA Broadcast-style microphone cleanup on Linux. It uses NVIDIA's Audio Effects SDK and sends the processed audio back through PipeWire as a virtual microphone.
This is an independent project, not an official NVIDIA application. I am sharing it early so other Linux users can try it and help me find what still needs fixing.
What it currently does
- Noise removal and BNR 2.0
- Room echo removal
- Combined noise and echo removal
- Low-latency Studio Voice
- Creates a microphone that can be selected in Discord, OBS, browsers and other Linux apps
The packages include the models and the CUDA, TensorRT and AFX libraries the app actually uses. You do not need to install a separate CUDA toolkit.
What you need
- A supported NVIDIA RTX GPU from the RTX 20, 30, 40 or 50 series
- An x86_64 Linux installation
- The proprietary NVIDIA driver
- PipeWire and a WebKitGTK 4.1-capable desktop
GTX, AMD, Intel and ARM devices are not supported at the moment.
Download v0.2.0
Choose your GPU generation first, then choose the package used by your Linux distribution. Each download is about 2.8 GB because the required models and runtimes are included.
Current status
This is still a preview. I tested all five modes on an RTX 4080 SUPER. The RTX 20, 30 and 50 packages pass the same model, runtime and architecture checks, but I have not yet tested them on those physical GPU generations.
If you try it, please open an issue with your Linux distribution, GPU, driver version and whether each mode worked. Reports from RTX 20, 30 and 50 users would help the most. Bug reports and packaging fixes are also welcome.