Arm motion symmetry in conversation

Data-driven synthesis of human motion during conversational speech is an active research area with applications that include character animation, computer gaming and agents. Natural looking key to both perceived realism understanding any synthesised animation. Multi-modal body-motion data scarce limited, so it common augment real by mirroring the body pose double number training samples. This augmentation based on assumption a person’s gesturing not affected handedness reflected plausible. In this study, we explore validity evaluating reflective symmetry speaker’s arms exchanges. We analyse left right arm 36 subjects dyadic conversation present per-frame gestures. To identify temporal offsets caused presence leading hand, compute time lag between movements arms. perform nearest neighbour search test mirrored pose. also consider information theory examine gain from data. implement speech-to-gesture generative model determine efficacy lateral techniques for augmentation. Our findings suggest positional left–right vary speaker speaker. conclude valid in certain cases when considering as new virtual identity, but should be carefully considered generic approach if style original maintained.