BAYESIAN METHOD FOR DETECTING EMOTION FROM VOICE FOR <i>KANSEI</i> ROBOTS

Abstract

The paper describes a Bayesian-based method for inferring a dialogist’s emotion from his or her voice, for Kansei robotics. The method uses a Bayesian model of prosodic features. This research focuses on the emotional elements included in the dialogist’s voice. Thus, the training datasets are prosodic features extracted from emotionally expressive voice data. The Bayesian information criterion, an information-theoretical selection method, is used in the learning Bayesian networks. Our method learns the dependence and its strength between the dialogist’s utterance and his emotional state. We propose a reasoner to infer the dialogist’s emotional state by using a Bayesian network for the prosodic features of the dialogist’s voice. The paper reports empirical reasoning findings and discusses how the relationship between certain components of prosodic features and certain emotions affects the reasoning.