VoxGen, an AMD-optimized TTS inference engine for VoxCPM 2 models
Heat trend
Collecting trend data
The percentage is based on available heat signal, not comment count or independent people.
Hi, everyone,
I’ve just released VoxGen, a lightweight native inference engine for VoxCPM2, written in Rust and using Vulkan compute instead of Python/PyTorch/CUDA.
Why VoxGen?
The main reason I started the project was because I needed a decent local text-to-speech solution.
I therefore saw VoxCPM 2 as a reasonable solution. However, most frameworks are NVIDIA-first, and VoxCPM 2 is no exception; as a result, my card was severely stuttering, and my GPU was always spiking. Also, having Python and Pytorch as a dependency is absolute hell.
This is why VoxCPM was created: not only we sidestep Pytorch completely, but performance on AMD cards is buttery smooth (and if you have a XTX 7900, I have designed a mode with even more aggressive power and speed optimizations)!
This application can also be run from a shell, so it can be integrated with other programs and scripts!
Installation:
You'll only need voxgen.exe (or the Linux equivalent) and the following files at https://huggingface.co/DennisHuang648/VoxCPM2-GGUF:
VoxCPM2-BaseLM-Q8_0.gguf VoxCPM2-Acoustic-F16.gguf
And that's it!
If you are interested, check out the Github page: https://github.com/NullMagic2/VoxGen Prebuilt binaries (for now, Windows only) are available here: https://github.com/NullMagic2/VoxGen/releases