An artificial neural network approach to modeling speech production is presented. The model uses simultaneously recorded data for neuromuscular activity (EMG), articulator motion, and the speech acoustics and based on dynamic optimization principle based on forward dynamics of articulators. The experimental configuration and data processing are explained, then the simulation result of forward dynamics modeling, which is the essential part of our modeling approach, is presented. Preliminary result of the forward acoustic modeling, which learns the correlation between articulator trajectories for the lips & jaw and PARCOR parameters of acoustic waveform, is presented.