lunavox 2.1.0

LunaVox tooling CLI for model setup, conversion, quantization, and build workflows.

🌌 LunaVox: High-Performance C++ Inference Engine for Qwen3-TTS

LunaVox is a high-performance C++ inference engine specifically designed for Qwen3-TTS. Through streamlined architecture and deep hardware optimization, it provides extreme speech synthesis speed and flexibility.

---

🚀 Key Features

• Lightweight Runtime: Runs with only ONNX Runtime and a custom Llama inference library, no heavy Python environment required.
• Native Multi-language Support: Built-in automatic language detection, supporting 10 languages.
• Full Mode Support: Supports Base synthesis, Voice Cloning, Custom Voice, and Voice Design.
• Cross-platform Hardware Acceleration: Deeply integrated with CUDA, CoreML/Metal, DML, and Vulkan.

---

🛠️ Requirements

• Python 3.10+: For model management.
• CMake 3.16+: For building the engine.
• Environment: Windows 10+, Linux, or macOS 12+.

---

📊 Performance

Configuration	Average RTF	Relative Speedup
LunaVox (Full CPU)	1.152	4.40x
LunaVox (CUDA13)	0.254	19.94x
LunaVox (Vulkan + DML)	0.206	24.59x

---

📦 Quick Start

Install via PyPI

pip install lunavox

Auto Setup

lunavox bootstrap

---

🎙️ Basic Usage

./build/qwen3-tts-cli.exe `
  -m models/base_small `
  -t "Hi, this is lunavox speaking English." `
  -o output/out.wav

---

📜 Documentation

• English Documentation Hub
• CLI Reference Guide
• Performance Report
• Runtime Tech Specs

---

🙏 Acknowledgements

• Qwen3-TTS
• onnxruntime
• llama.cpp