Laughter animation generation

Abstract

Laughter is an important communicative signal in human-human communication. It involves the whole body, from lip motion, facial expression, to rhythmic body and shoulder movement. Laughter is an important social signal in human-human interaction and may convey a wide range of meanings (extreme happiness, social bounding, politeness, irony,). To enhance human-machine interactions, efforts have been made to endow embodied conversational agents, ECAs, with laughing capabilities. Recently, motion capture technologies have been applied to record laughter behaviors including facial expressions and body movements. It allows investigating the temporal relationship of laughter behaviors in details. Based on the available data, researchers have made efforts to develop automatic generation models of laughter animations. These models control the multimodal behaviors of ECAs including lip motions, upper facial expressions, head rotations, shoulder shaking, and torso movements. The underlying idea of these works is to propose a statistical framework able to automatically capture the correlation between laughter audio and multimodal behaviors. In the synthesis phase, the captured correlation is rendered into synthesized animations according to laughter audio given in input. This chapter reviews existing works on automatic generation of laughter animation.