Multi-party Interactions by Quizmaster Robot in Speech-Based Jeopardy! Like Games

摘要

Robot Audition enables a robot to listen to simultaneous utterances via its own ears by localizing and separating sound sources and recognizing separated sounds. Although several applications are developed as a proof of concept, such capabilities of robot audition have not been well shaped in the context of applications, particularly, in multi-parson interactions. This paper focuses on the question "what is the next step when a robot and/or system can listen to several utterances at once?'' As an example of multi-party interactions, the paper describes a quizmaster robot for speech-based Jeopardy! like games with two interaction models, school-class-type and auction-type. The player of the quiz answers a question by getting the right to answer in the former, while he/she can say an answer directly. Empirical evaluation of the system and lessons are discussed.