遗传算法与BP神经网络相结合的说话人识别系统

Human speaker recognition based on integration of genetic algorithm and BP neural network

基于BP神经网络的说话人识别系统是目前说话人识别中的一种主要模型,但BP神经网络通常难以确定隐含层单元的数目,且收敛速度慢。针对此缺点,提出了一种基于遗传算法(GA)的说话人识别BP神经网络优化方案,该方案利用混合编码的GA对神经网络的连接权和结构进行了优化,可以有效地剔除整个网络冗余节点和冗余连接权,方案利用了BP神经网络的并行性和GA的全局搜索能力,显著地改善了网络的处理能力。实验表明:基于混合编码GA的BP神经网络具有快速学习网络权重的能力,识别率高,是说话人识别的一种有效可行的新方案。

Although the human speaker recognition system based on back propagation（BP） neural networks is one of the main models for distinguishing speaker,it has some shortcomings ,this is,the hidden layer node number of BP network is often hard to be assigned, and the system has a slow convergence speed. A BP neural networks optimization scheme based on genetic algorithm（GA） for human speaker recognition is proposed. In the scheme, the hybrid encoding GA is used to optimize the connected weights and structure of BP neural networks and the redundant nodes and redundant connected weights are removed from the networks effectively. The scheme utilizes the parallelism of the neural networks and the global search capability of the GA, so it improves the processing capability of the networks evidently. The experimental tests show that the scheme based on hybrid encoding GA has a fast learning speed, a high recognition rate, and it is a new practical scheme for human speaker recognition.