A Study of User's Acceptance Evaluation Model for Paralinguistic Information of Response Speech in Spoken Dialogue System

Abstract

Today, Japan's super-aged society and other social backgrounds are increasing the demand for dialogue partners. Spoken dialogue systems are expected to be utilized for this demand. One of the roles of a dialogue system as a chat partner is to share mental states. When a dialogue system responds to a user's expressed mental state with paralinguistic information that the user cannot tolerate, the dialogue breaks down and the user's desire to continue the dialogue decreases. This research aims to solve the problem of such a breakdown. We have focused on the pleasant and unpleasant states expressed by user utterances and system responses, and have confirmed the existence of an acceptable range of system responses to various user utterances. In this paper, we discuss a tolerance evaluation model for outputting acceptable system responses according to the pleasant and unpleasant states expressed in user speech.