A Method Of Design And Embedded Implementation Of Speaker Recognition System Based On BP Neural Network

The identification of speaker is a kind of biometric authentication depending on the voice. At present, the identification is mostly based on traditional linear system theory. But the identification of speaker itself is a complicated non-linear classification process. Based on the non-linear distributed parallel-processing network pattern, artificial neural network has strong ability of pattern classification and robustness to the incomplete information. The research on the identification based on neural network is a inevitable tendency, which provides a unique way and thought to the identification of speaker.The way of voice signal preprocessing and endpoints detection and picking up features is introduced in detail in this paper. At the same time the system by using BP neural network as classifier is also set up. To the question that BP Neural Network has slow velocity of convergence and is apt to fall into the minimum value, chaos thought is adopted in the particle swarm optimization algorithm. For this, chaos-particle swarm optimization algorithm, which improve the ability of getting rid of fractional extreme point in the particle swarm optimization algorithm, is presented and applied to the BP network exercise so that the calculation accuracy and velocity of convergence of BP network are increased. By putting the BP network exercise into the identification of speaker, the discrimination and training speed are remarkably increased and identification of speaker based on BP Neural Network also gets better effect.In addition, boot program is completed based on the ARM and Linux embedded platform. The root file system is set up by transplanting and tailoring linux kernel and finally the program of identification of speaker is successfully transplanted into embedded system.