基于语音信号的时变特性，研究了神经网络语音识别的方法．把混沌特性引入到神经元，构造了一种新的多层混沌神经网络结构，同时推导了相应的学习算法．把这种混沌神经网络用于语音识别，并与常用的神经网络语音识别方法作了比较．实验结果表明，混沌神经网络方法的平均识别率要高于同等条件下常用神经网络方法的识别率． Based on the time-varying characteristics of speech signals, the method of neural network speech recognition is studied. By introducing the chaotic characteristics into neurons, a new structure of multi-layer chaotic neural networks is constructed and the corresponding learning algorithms are derived. This chaotic neural network is used for speech recognition and compared with the commonly used neural network speech recognition method. Experimental results show that the average recognition rate of the chaotic neural network is higher than that of the commonly used neural networks under the same conditions.