NaruseMioShirakana/MoeVoiceStudio
Free一个使用C++编写的音频处理软件
About NaruseMioShirakana/MoeVoiceStudio
MoeVoiceStudio is a branch of the DragonianVoice project, providing a simple Qt-based graphical user interface for the DragonianVoice ONNX inference library. DragonianVoice is an open-source, offline C++ inference framework for various voice synthesis models, including Text-to-Speech (TTS) models like Tacotron2, Vits, EmotionalVits, BERTVits2, and GPtSoVits; Singing Voice Conversion (SVC) models like SoVitsSvc, RVC, DiffusionSvc, FishDiffusion, and ReflowSvc; and Singing Voice Synthesis (SVS) models like DiffSinger. It supports C, C++, and C# API calls, uses ONNX Runtime for inference, and runs completely offline without collecting any user data. The project is free, open-source, and community-driven, but note that the UI/maintenance for this branch has been deprecated in favor of a pure library approach.
Key Features
Pros & Cons
- Free and open-source with no hidden costs
- Supports a wide range of state-of-the-art voice models
- Runs entirely offline, ensuring user privacy
- Cross-platform with Qt GUI and C++ core
- Active community with regular updates (as of project state)
- Provides C, C++, and C# APIs for integration
- Requires conversion of PyTorch models to ONNX format first
- Technical knowledge needed for setup and model preparation
- GUI branch (MoeVoiceStudio) no longer actively maintained; focus shifted to pure library
- Some models (e.g., SoVits 3.0/4.0) had issues with DirectML on older ONNX exports
- Diffusion/Reflow models can produce noisy output if inference steps exceed training max steps