Hello!
It's been a rough week of constant patches, but a major breakthrough was made. I found a way to make Qwen3 Voice Design / Reference Voice and AI Assistance significantly faster.
These features no longer require the GPU Acceleration package. You don't need to install it for them anymore.
Changes
Moved Qwen3 Voice Design / Reference Voice and AI Assistance to a native C++ implementation, providing much faster performance with CUDA and Vulkan.
Added the missing CUDA package for Chatterbox, Resemble Enhance, and RVC, allowing CUDA users to run them much faster. (Optional via the GPU Acceleration package.)
Updated the Piper voice catalog with the latest upstream voices.
Added Chatterbox Nano, a smaller and faster alternative to Chatterbox Turbo.
Fixed a wide assortment of bugs and stability issues.
Improved UI responsiveness throughout the application, especially in Settings.
Replaced the previous Kokoro Chinese package with a better pre-packaged solution. This accounts for most of the increased download size.
Preview Branch
Text to Music / Sing has arrived in the Preview branch. It's still early but it's a really exciting step forward.