k2-fsa/OmniVoice

Why this repo matters

• •Supports zero-shot voice cloning across 600+ languages, offering the broadest multilingual coverage currently available.
• •Achieves 40x real-time inference speed using a novel diffusion language model architecture for efficient production deployment.
• •Enables fine-grained voice design and non-verbal control through speaker attributes and phoneme-level pronunciation correction.