Figure 3From: Model-Based Synthesis of Visual Speech Movements from 3D Video Schematic of the synthesis process: stored phoneme exemplars along with the input audio features are used to select optimal units to train a state-based manifold traversal model. Back to article page