Research on autonomous interaction models of virtual avatar

When interact through avatar-based virtual systems, remote users can communicate each other with vivid feelings in virtual face-to-face means. Usually actions of a virtual avatar are controlled by its master´s operations on the system. If a user doesn´t operate the system for a long period, his/her avatar cannot indicate the real intentions and emotions to the interaction partner, because the avatar falls into action blind period. The interaction models among persons in daily life are analyzed. Operation frequency of the user on virtual system can indicate the interest extent of the user. An adaptive time window based algorithm for judging users becoming operation silence or not is proposed. When the user loses interest of topic, he/she appears sleepy states. According to different silence extent, virtual avatar is matched with suitable sleepy models. This research is applied to a 3D communication system based on text-to-visual speech synthesis (TTVS). The results show that the proposed model maintains the continuity of the virtual interaction system.