The KTH talking head in space - a vehicle for situated multi-party interaction

Prof. Bjorn Granstrom

KTH Royal Institute of Technology, Stockholm

The KTH 3D talking head model has been used for more than ten years. It has been applied mostly in spoken dialogue systems and as a lip reading support for hard of hearing persons. In all cases the 3D model has been displayed on 2D computer displays. Recently we have experimented with the same model displayed on 3D facial masks, like the back-projected solution incorporated in a robotic head. This presentation will describe some of the new opportunities and challenges when moving from flat screens to a 3D rendering in the physical space of the user. While the original ambition with the model was to accurately display segmental articulation, we have recently expanded the capabilities of the head to prosodic and non-verbal signals. Several studies concern aspects of human-robot interaction in a multi-party setting, where e.g. selective gaze control is of great importance. Preliminary results from a comparison of our back projected head with a more conventional mechatronic robot head will also be presented as part of the EU IURO project.