Quantifying Sign Avatar Perception: How Imperfect is Insufficient?

The aim of this study was to identify relevant aspects of sign avatar animation and to quantify their effect on human perception, to better address user expectation in the future. For this, 25 users assessed two types of avatar animations, one upper baseline generated utilizing motion capture data, and a reference with absolute total positional differences in the mm range generated utilizing machine learned synthetic data. As expected, user evaluation of the synthesized references showed a considerable loss in rating scores. We therefore computed a variety of signal-specific differences between both data types and investigated their correlations to the collected user ratings. Results reveal general statistically significant inter-dependencies of avatar movement and perception helpful for the generation of any type of virtual avatar in the future. However, results also suggest that it is difficult to determine concrete avatar features with a high influence on the user perception in the current study design.