Virtual Listener: A Turing-like test for behavioral believability

Abstract

Abstract Virtual Listener (VL) is a generalized prototype of a virtual character based on the principles of cognitive architecture eBICA, which uses facial expressions and “body language” (eyes movements, head rotation) to keep social and emotional contact with the user. Such contact also implies that VL needs to perceive user’s facial expression and gaze, and in the long term – also intonation of the user’s voice, the sentiment and content of user’s speech. In this work, we present an approach to modeling a perceptive 3D virtual listener with emotional capabilities. The virtual character has a 3D face that performs real-time, realistic and believable facial expression dynamics. Our primary goal in this study was to evaluate the concept: e.g., to find out whether a virtual-agent-generated behavior can engender feelings of rapport in human speakers comparable to those that a real human listener can cause? At the same time, this article serves a limited purpose and only describes our current progress so far in addressing this question.