Aivis-Project/AivisSpeech-Engine

# Technical Summary Built on VOICEVOX ENGINE's architecture, this engine uses ONNX Runtime for efficient CPU-based inference while supporting the AIVMX model format (.aivmx)—a specialized container combining learned models, hyperparameters, style vectors, and speaker metadata. The REST API maintains VOICEVOX compatibility but extends functionality with emotion-rich voice generation and style control through embedded style IDs. Deployable across Windows, macOS, and Linux with optional NVIDIA GPU acceleration via CUDA or DirectML, it integrates with the AivisSpeech desktop application and supports model discovery via local directories or S3-compatible storage.

150 stars.