API Documentation#
Streaming APIs#
The system provides three versions of streaming APIs with different capabilities and deployment strategies:
V1 API: Basic streaming motion generation interface with fundamental motion keyword matching. This version is currently deprecated and no longer actively maintained.
V2 API: Enhanced streaming interface built upon V1’s foundation, featuring:
Advanced retrieval capabilities for emotions and relationship annotations
Enhanced support for diverse downstream applications
Improved interpolation and memory management
V3 API: Revolutionary streaming interface with a completely different dual-timeline synchronous retrieval strategy:
Significantly increased keyword motion trigger rates
Advanced blending capabilities and improved transitions
Dual-timeline architecture for enhanced motion synchronization
Deployment: Both V2 and V3 APIs are simultaneously deployed and accessible through different FastAPI service endpoints, allowing applications to choose the most suitable version based on their specific requirements.
Request/Response Format#
All APIs use Protocol Buffers for efficient serialization. The system supports:
Chunk-based Processing: Speech input is processed in chunks for real-time response
Motion Timeline: Frame-based timeline management for precise motion sequencing
Keyword Matching: Both motion keywords and speech keywords for intelligent selection
Memory Integration: User session memory to avoid repetitive animations