Design And Implementation Of Speaker Recognition System Based On ARM

At present, the speaker recognition system generally uses the PC or server as a platform, which inevitably exists in the problem of large size, high power consumption, low portability and practicality, leading to the unpopularity. Aiming at this problem, a portable speaker recognition system based on ARM is designed and implemented.In the software algorithm, the basic algorithms of speaker recognition system are analyzed and then improved. In this paper, an adaptive endpoint detection algorithm based on time-frequency parameter is proposed, which improves the accuracy and noise immunity of endpoint detection, solving the problem that the common endpoint detection algorithm can not effectively segment words in noisy environment. Otherwise, a feature extraction algorithm based on the Bark wavelet packet transform with Fisher is proposed, which obviously enhances the recognition rate and robustness of the whole system, solving the problem of low recognition rate of MFCC parameter.In the design and implementation of system platform, according to the demands of the software algorithm and system function, a development board called OK6410 is used as the system hardware platform, then Linux operating system as the system software platform is transplanted. On this basis, the overall architecture of the system is designed to realize the speech acquisition and playback function, design the software algorithms, accomplish the application design through user interface by QT, and test the whole system.The test shows that the system has achieved the expected design goal and owns good portability, accuracy and robustness. It provides a new solution for the password payment in real life.