Game Development Reference
In-Depth Information
animation was restricted to cartoon-like figures, which do not show the same
complexity as real faces). Our selection came out to be a good compromise
between the number of visemes needed in the animation and the realism that is
obtained.
It is important to note that our speech model combines the visemes with additional
co-articulation effects. Further increases in realism are also obtained by adapting
the viseme deformations to the shape of the face. These aspects are described
in the section,
Face Animation
.
Learning Viseme Expressions
The deformations that come with the different visemes had to be analyzed
carefully. The point of departure in developing the animation system has,
therefore, been to extract detailed, 3D deformations during speech for ten
example faces. These faces differed in age, race, and gender. A first issue was
the actual part of the face that had to be acquired. The results of Munhall and
Vatikiotis-Bateson (Munhall et al., 1998) provide evidence that lip and jaw
motions affect the entire facial structure below the eyes. Therefore, we
extracted 3D data for a complete face, but with emphasis on the area between
the eyes and the chin. The extraction of the 3D visemes follows a number of
steps, which were repeated for the different example faces:
1.
a 3D reconstruction is produced for all instances of all visemes
2.
a generic head model is fitted to these 3D visemes
3.
prototypes of the visemes are defined
These steps are now described in more detail.
Raw Viseme Extraction
The first step in learning realistic, 3D face deformations for the different visemes
was to extract real deformations from talking faces. Before the data were
extracted, it had to be decided what the test person would say during the
acquisition. It was important that all relevant visemes would be observed at least
once. The subjects were asked to read a short text that contained multiple
instances of the visemes in Figure 2.
Search WWH ::
Custom Search