Converting emotional voice to motion for robot telepresence

In this paper we present a new method for producing affective motion for humanoid robots. The NAO robot, like other humanoids, does not possess facial features to convey emotion. Instead, our proposed system generates pose-independent robot movement using a description of emotion through speed, intensity, regularity and extent (DESIRE). We show how the DESIRE framework can link the emotional content of voice and gesture, without the need for an emotion recognition system. Our results show that DESIRE movement can be used to effectively convey at least four emotions with user agreement 60-75%, and that voices converted to motion through SIRE maintained the same emotion significantly higher than chance, even across cultures (German to Japanese). Additionally, portrayals recognized as happiness were rated significantly easier to understand with motion over voice alone.