ReCell: replicating recurrent cell for auto-regressive pose generation

Vladislav Korzun, Anna Beloborodova, Arkady Iliin

Track: Challenge paper

Team Name: FineMotion

Keywords: embodied agents, neural networks, gesture generation, social robotics, deep learning

Abstract: This paper describes FineMotion’s gesture generating system entry for the GENEA Challenge 2022. Our system is based on auto-regressive approach imitating recurrent cell. Combined with a special windowed auto-encoder and training approach this system generates plausible gestures appropriate to input speech.