Generating Personalized Virtual Agent in Speech Dialogue System for People with Dementia

Abstract

Our research group has been studying a speech communication system with a virtual agent (VA), to support person-centered care (PCC) of people with dementia (PWD). The current system uses the 3D model based on an unreal character for the VA. Because the unfamiliar appearance is to be a mental obstacle to PWD, PWD hardly accept advice and which causes a limitation in the care effects. In this paper, we develop a novel system that dynamically creates a VA based on a given facial image of real person. The proposed system constructs a three-dimensional model based on facial landmarks within the image. It then stretches and transforms some portions of the 3D model to generate facial expressions. From just a given picture, the proposed system easily generates a communication agent familiar with individual PWD. Hence, it can implement (virtual, but effective) conversations with familiar partners. We implement the prototype based on the proposed system and conduct the experiment targeting to the elderly.