Duplicated from  k2-fsa/OmniVoice

edwixx

/

OmniVoice

like 1

This pull request improves the model card for OmniVoice by adding detailed documentation from the research paper and official repository.

Key changes include:

• A summary of the model's architecture (Diffusion Language Model-style discrete NAR) and its capabilities (600+ languages, 581k-hour training data).
• Detailed installation instructions for both pip and uv.
• Functional Python code snippets for the primary use cases: Zero-shot Voice Cloning and Voice Design.
• Documentation of expressive features like non-verbal symbols (e.g., [laughter]) and pronunciation control.
• Proper citation for the OmniVoice paper.

These additions make the repository much more accessible to researchers and developers looking to utilize the model.