Reproduce “MoSAT: Human Motion Generation from Spatial Audio and Textual Description” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shuyang Xu, Zhiyang Dou, Yiduo Hao, Zekun Li, Liang Pan, Jingbo Wang, Cheng Lin, Yuan Liu and 3 more
MoSAT combines language for action intent and style with spatial audio for event timing and direction. It encodes skeletal motion with a Motion VAE, uses CLAP features and text plus initial-frame DOA context, and generates motion with Transformer flow-matching. On STAM, it reports FID 0.73, CMR R@1 37.05, and MCR R@1 37.95, improving retrieval over EDGE but not isolating audio’s effect.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.