抄録
Voice interaction systems and communication systems with dialog agents using speech recognition have been proposed and developed. On the other hand, a speech-driven embodied entailment CG character called "InterActor" has been developed, which automatically generates communicative motions and actions such as nods for entrained interaction from voice rhythm based on only speech input. However, conventional InterActor generates only positive actions for all of the utterance content but, doesn't generate negative actions. Accordingly, there is a possibility to promote talker's negative feelings by performing positive actions to negative utterance content. In this paper, we develop an advanced speech-driven embodied entrainment character system with response to improve talker's feelings by using speech recognition. In this system, talker's utterance is converted to text by speech recognition, and talker's emotion is estimated from character strings in the converted text. It uses a database that quantifies each word in the estimation of emotion. Then, the system automatically generates embodied entrainment motions or soothing and activated actions corresponding to emotion.
