Mitigating Error Accumulation in Co-Speech Motion Generation via Global Rotation Diffusion and Multi-Level Constraints

Xiangyue Zhang, Jianfang Li, Jianqiang Ren, Jiaxu Zhang · 2025

Read Mitigating Error Accumulation in Co-Speech Motion Generation via Global Rotation Diffusion and Multi-Level Constraints by Xiangyue Zhang, Jianfang Li, Jianqiang Ren, Jiaxu Zhang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-10076.txt.

Description

Xiangyue Zhang, Jianfang Li, Jianqiang Ren, and Jiaxu Zhang generate gestures that move along with speech, over long stretches, without falling apart. Most generators work with local joint rotations defined through the skeleton's hierarchy, so small errors stack up along each limb and end effectors drift into implausible poses. Their framework, GlobalDiff, runs diffusion directly in the space of global joint rotations instead.

Working globally decouples each joint from upstream mistakes, but it removes the skeleton's built-in structural guidance. Three constraint levels restore it: virtual anchors around joints for fine orientation, angular consistency across bones for integrity, and a multi-scale temporal encoder for rhythm. On standard co-speech benchmarks the system improved performance by about forty-six percent over the previous best across several speakers.

Catalog record

CREATORXiangyue Zhang, Jianfang Li, Jianqiang Ren, Jiaxu Zhang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-10076

Browse the full library · About the Tower of Babel