基于改进的神经网络异常声音自动识别系统研究

针对标准的BP神经网络对于声音信号识别率不高的问题，提出了一种用粒子群算法（PSO）优化BP神经网络的算法，建立了声音信号识别模型。PSO优化BP神经网络主要是用PSO来优化BP神经网络的初始权值和闽值，然后通过训练BP神经网络得到识别模型的最优解，优化后的神经网络具有误判率小、反应速度快等特点。在实验中把标准的BP神经网络和PSO优化后的BP神经网络用于八种异常声音的MFCC特征量和差分MFCC特征量识别，结果表明：在声音信号的识别系统中采用PSO优化BP神经网络的算法提高了系统的识别性能，达到了系统设计的目的。

Abnormal Sound Automatic Recognition System Based on Modified Neural Network

According to the low recognition rate of the standard BP neural network, a neural network based on the particle swarm optimization algorithm is put forward and a sound recognition model is established. PSO is mainly used to optimize BP neural network＇s initial weights and threshold value,obtain the optimal solution of recognition model through training BP neural network by BP neural network optimized by PSO and the optimized neural network have a relatively low error rate and a relatively fast speed etc. In the experiment the MFCC characteristic value and the difference MFCC characteristic value of eight kinds of abnormal sound based on BP neural and optimized BP neural network is recognized; The results show that the system,s recognition performance through using the BP neural network optimized by PSO is improved,and the purpose of the system design is achieved.