Motion Generation Of Conversational Character From Labeled Script

Abstract

In order to realize a character that can talk with voice, it is necessary to be able to control the transition of poses across multiple utterances and the expression of emotions according to the dialogue. In this paper, we propose a method for generating customer actors' performances that can represent various situations. We extract labels related to the conversation scenario and synthesize body motions, gestures, and head/eye motions of the customer actor according to different situations corresponding to different labels. We discuss the results of generating acting motions using a dialogue scene between a staff member and a customer in a customer service situation as a concrete example.