Towards Multimodal Expression of Laughter

Laughter is a strong social signal in human-human and human-machine communication. However, very few attempts to model it exist. In this paper we discuss several challenges regarding the generation of laughs. We focus, more particularly, on two aspects a) modeling laughter with different intensities and b) modeling respiration behavior during laughter. Both of these models combine a data-driven approach with high-level animation control. Careful analysis and implementation of the synchronization mechanisms linking visual and respiratory cues has been undertaken. It allows us to reproduce the highly correlated multimodal signals of laughter on a 3D virtual agent.