PriesiaMioShirakana/DragonianVoice

Unified ONNX-based inference library supporting 13+ voice synthesis models (TTS, SVC, SVS) with C/C++/C# bindings and multi-backend acceleration (CUDA, DML, CPU). Implements model-agnostic architecture handling Tacotron2, VITS, RVC, DiffSinger, and diffusion-based variants through standardized JSON configuration and prerequisite models (HuBERT, HiFiGAN). Provides command-line and library interfaces with pitch prediction (FCPE/RMVPE) and step-count validation for diffusion models to prevent audio artifacts.

1,121 stars. No commits in the last 6 months.