Having at our disposal accurate phoneme segmentation and spatio-temporal correspondences among all facial scans, we can arrange the 3-D face scans into groups corresponding to particular phonemes, and thus build a statistical model of the phonemes’ visual appearance (visemes).