Using Timing Information to Improve the Performance of Avatars

Thomas Hanke, Silke Matthes, Anja Regen, Jakob Storz, Satu Worseck, Ralph Elliott, John Glauert and Richard Kennaway


In recent years, much work has been done to improve the performance of avatars used to generate continuous signing from notation. However, while much effort has been spend on the manual performance of the signing avatars, especially on hand configurations, comparatively little attention has been paid to rhythm in signing. Sign language produced by an avatar driven from symbolic description of sign sequences is often reported as unnatural or boring, or difficult to segment, and our belief is that improvements on the avatar's signing rhythm would mean a significant step in making avatar movements look more natural and be easier to interpret. We present a small-scale study that provides detailed timing information for individual sign parameters in natural signing as well as the tools used to collect this data with modest costs. The avatar system is being improved to be able to play back signing annotated in a combination of symbolic movement description and precise timing information. We will report on microtesting how much this actually improves subjective recognisability. In a second step, we generalise from the available data set and apply per-parameter timing patterns to animation from notation alone. Once again, microtesting will show if this improves avatar recognisability.




Requirements for signing avatar technology
Realism and acceptability of signing avatars
Linguistically-informed notations for gestural animation
Use of corpora to inform animation
Realistic animation of manual and bodily gestures


[Paper (PDF)]  

Supported by START Conference Manager