Human-Machine Interaction Compensating Imperfection in Facial Expression Recognition

This study aims at comfortable communication between humans and machines. We propose an interactive voice output system for human-machine interaction. Facial expression is focused on and automatically recognized as one of four expression candidates: "joy," "anger," "sadness," "surprise." The facial expression recognition uses features that correspond to expression-motion-patterns. The first and second expression candidates are chosen from results of the facial expression recognition. Information on these candidates is used for accessing our voice database. The voice database contains voice data corresponding to human feelings. In order to realize the interactive voice output system, the voice data are outputted with compensating imperfection in the facial expression recognition, since perfect facial expression recognition has not been developed yet.