Knowledge sources in spoken dialogue systems

摘要

This paper describes a robot who converses with multi-person using his multi-modal interface. The multi-person conversation includes many new problems, which are not cared in the conventional oneto-one conversation: such as information flow problems (recognizing who is speaking and to whom he is speaking / appealing to whom the system is speaking), space information sharing problem and turn holder estimation problem (estimating who is the next speaker). We solved these problems by utilizing multi-modal interface: face direction recognition, gesture recognition, sound direction recognition, speech recognition and gestural expression. The systematic combination of these functions realized human friendly multi-person conversation system.