EVA: expressive multipart virtual agent performing gestures and emotionsMlakar, Izidor (Avtor) Rojc, Matej (Avtor) bone and morph based animationdistributiveexpressive ECAmesh based articulated modelEmbodied Conversational Agents (ECAs) play an important role in the development of personalized and expressive human-machine interaction, allowing users to interact with a system over several communication channels, such as: natural speech, facial expression, and different body gestures. This paper presents a novel approach to the generation of ECAs for multimodal interfaces, by using the proprietary EVA framework. EVAćs articulated 3D model is mesh-based and built on the multipart concept. Each of its 3D submodels (body-parts) supports both bone and morph target-based animation, in order to simulate natural human movement. Each body movement's structural characteristics can be described by the composite movement of one or more elementary units (bones and/or morphs), and its temporal characteristics by the durations of each of the movementćs stages (expose, present, dissipate). EVA scripts provide a means of defining and fine-tuning body motion in the form of predefined gestures, or complex behavioural events (provided by external behaviour modelling sources). Since behavioural events can also be described as a combination of tuned predefined gestures and the movements of elementary units, a small number of predefined gestures can form infinite sets of gestures that ECA can perform. ECA EVA, as presented in this paper, provides both: a personalization of its behaviour (gesture level), and a personalization of its outlook.20112012-06-01 10:26:35Neznano27297UDK: 004.94COBISS_ID: 14648342ISSN pri članku: Y506-2586NUK URN: URN:SI:UM:DK:YO6ELZSXsl