Speech production is probably the most complex in various human motion control tasks. In the current neural network model of speech production ,DIVA model is relatively the most thorough , and is a unique pseudo- inverse model control scheme . This control scheme provides a very accurate interpretation of the kinematic data of the human vocal organs .Because the speech recognition process is too complicated , so recognition process in the current DIVA model of the system is just an expert system .The paper first reviews the development process of speech production model , describes the development and current status of the speech recognition technology . The DIVA speech production model is then introduced , including the basic concepts of the DIVA model , the structural framework , the status and inadequacies of the model. Then discusses processes and methods of pre-processing and feature extraction for speech recognition . In order to explain the phenomenon of perceptual magnetic effect ,it proposes a neural model based on self- organizing feature map ,and analysis the feasibility and correctness of the model by computer simulation. Speech recognition system in DIVA model is only a expert system ,so it improves the algorithm of the model proposed above and apply it to the DIVA model.The Simulation experiment shows that we reach the goal. Finally , It summarizes the paper , points out the inadequacies of current research , and pointed out the follow-up research directions for further research. |