Authors:
Afef Cherni
1
;
Roxane Bertrand
2
and
Magalie Ochs
3
Affiliations:
1
Aix-Marseille Univ., IN2P3, CNRS, France
;
2
Aix-Marseille Univ., LPL, CNRS, France
;
3
Aix-Marseille Univ., LIS, CNRS, France
Keyword(s):
Multimodal Cues, Persuasion, Embodied Conversational Agent, Machine Learning Methods, Mathematical Convolution.
Abstract:
The persuasiveness of a virtual agent refers to its ability to influence, persuade, or motivate users to take specific actions or adopt certain attitudes or beliefs. Virtual agents can use its multimodal capabilities, including non-verbal cues to enhance their persuasiveness. In this paper, we present a new tool called THRUST (from neuTral Human face to peRsUaSive virTual face) to automatically generate the head movements and facial expressions of a persuasive virtual character. This tool is based on a machine learning approach from a human videos corpus to identify the non-verbal persuasive cues. A convolution-based model then transforms neutral non-verbal behavior to a persuasive non-verbal behavior simulated on a virtual face. Videos generated by the tool have been evaluated through a subjective perceptive study with about 90 participants. The results show that the virtual agent’s head and facial behaviors generated by the THRUST tool are perceived as persuasive, thus validating t
he proposed approach.
(More)