TaoMate: Anchor-Guided Memory Bridging Evolving and Reference States for Real-Time Audio-Video Digital Human Generation

Researchers introduced TaoMate, an anchor-guided memory framework for long-form real-time audio-video digital human generation.

It compresses past audio-video blocks into reusable memory states to support causal generation over extended sequences.

Residual attention and reference-aware modulation help preserve appearance and audio-video synchronization.

The approach also enables faster stage-parallel inference, improving stability and efficiency for interactive media.