Hi-Reco
The whole picture of this project
Abstract:
High-fidelity digital humans are increasingly usedin interactive applications, yet achieving both visual realism and real-time responsiveness remains a major challengeWe present a high-fidelity, real-time conversational digitalhuman system that seamlessly combines a visually realistic3D avatar, persona-driven expressive speech synthesis, andknowledge-grounded dialogue generation. To support naturaland timely interaction, we introduce an asynchronous execu!tion pipeline that coordinates multi-modal components withminimal latency. The system supports enhancements suchas wake word detection, emotionally expressive prosody.and context-aware response generation through retrievalaugmented methods. Together, these components form anintegrated framework that enables responsive, believable digital humans suitable for immersive applications in communication, education, and entertainment.
For more details please check here .
Junwei Li
Research Fellow
My research interests include Multi-Agent LLM Systems, RAG/Graph-RAG, Educational Technology and compliance-aware AI companions for cloud–edge education and assistive use.