Conversational agent expressing ocean personality and emotions using laban movement analysis and nonverbal communication cues

Abstract

Conversational human characters are heavily used in computer animation to convey various messages. Appearance, movement and voice of such characters in uence their perceived personality. Analyzing different channels of human communication, including body language, facial expression and vocalics, it is possible to design animation that exhibit consistent personality. This would enhance the message and improve realism of the virtual character. Using OCEAN personality model, we design internal agent parameters that are mapped into movement and sound modi ers, which in turn produce the nal animation. Laban Movement Analysis and Nonverbal Communication Cues are used for the operations that output bone rotations and facial shape key values at each frame. Correlations between personality and spoken text, and relations between personality and vocal features are integrated to introduce compherensive agent behavior. Multiple animation modi cation algorithms and a personality based dialogue selection method is introduced. Resulting conversational agent is tested in different scenarios, including passport check and fastfood order. Using a speech to text API user controls the dialog ow. Recorded interactions are evaluated using Amazon Mechanical Turk. Multiple statements about agent personality are rated by the crowd. In each experiment, one personality parameter is set to an extreme while others remain neutral, expecting an effect on perception.