Basic research to a virtual face-to-face communication environment between an operator and a machine is presented. In this system, human natural face appears on the display of machine and can talk to operator with natural voice and natural face expressions. Especially in this paper, face expression synthesis scheme driven by natural voice is presented. Voice is including not only linguistic information but also emotional features. So we proposed expression control scheme driven by both features. We express a human head with 3D wire frame model. The surface model is generated by texture mapping with 2D real image. All the motions and expressions are synthesized and controlled automatically by the movement of some feature points on the model.