audio.cpp (CUDA)
audio.cpp contributors
Run local audio AI models with the NVIDIA CUDA 12.8 backend
Copied! Paste it into a terminal and press Enter to install.

About
Run speech synthesis, speech recognition, voice conversion and music generation locally with audio.cpp. This community package uses the unmodified official CUDA 12.8 Linux server release.
The launcher starts a local server in a terminal; the package has no native application window. Open the loopback address 127.0.0.1 on port 8083 in your browser to use its embedded WebUI. Close the server with Ctrl+C. Models and voices are stored in this app's private data directory. Model weights have their own licenses.
This package includes the native model manager in the server. Workflows requiring Python conversion or additional host tools are unavailable. CPU, Vulkan and CUDA editions use separate model stores and ports. This edition requires a compatible NVIDIA GPU and driver; Flatpak supplies the matching driver extension.
The upstream CUDA Linux archive contains the server only; the standalone CLI is not included. To access an existing models folder, grant it with flatpak override --user --filesystem=/path/to/models:ro io.github._0xShug0.AudioCpp.CUDA. Undo custom grants with flatpak override --user --reset io.github._0xShug0.AudioCpp.CUDA.
Details
- Developer
- audio.cpp contributors
- License
- Apache-2.0
- Runtime
- Freedesktop 26.08
- Architectures
- x86_64
- Version
- 0.9.6 · 2026-11-21
- Website
- Homepage
- Source
- audio.cpp (CUDA)
- Packaging
- flatpark/registry
Permissions
2 to reviewAll devices (incl. cameras, USB)
Network access
Can reach the internet and local network
Packaging
AdaptedAdds components the runtime lacks
Shipped next to the app: nccl