基于动态优化神经网络的非确定语音识别研究  

Speech Recognition is Determined Based on the Dynamic Optimization of Neural Network Research

作　　者：马建红[1,2] 姬莉霞[1,2] 张晗[2] 

机构地区：[1]武汉理工大学信息工程学院,武汉430070 [2]郑州大学软件技术学院,郑州450002

出　　处：《计算机测量与控制》2013年第7期1996-1998,共3页Computer Measurement &Control

基　　金：河南省科技攻关项目(122102210518);河南省教育厅科学技术研究重点项目(12A520042)

摘　　要：语音识别面对不同类型对象,传统的线性系统处理复杂度高的非特定语音准确率较低,提出一种基于优化神经网络的家电控制语音识别方法,将非特定的语音时域波形的分形维数作为神经网络的语音识别的特征参数,结合传统的语音特征参数构造语音识别的神经网络模型,在模型构造过程中使用粒子群优化的方法将神经网络的权重系数进行动态调整,防止优化模型陷入局部最小,实验证明,这种算法实现的家电控制的语音识别准确率高达92%,较传统的线性系统准确率提高了20%,具有很强的实用价值。Speech recognition in the face of the different types of object, the traditional linear system processing complexity high nonspe- cifie voice low accuracy, this paper presents a neural network based on optimization of home appliance control speech recognition method, will not specific speech time domain waveform fractal dimension as a neural network speech recognition characteristics parameters, and combining with the traditional speech feature parameters structure speech recognition neural network model, the model structure used in the process of particle swarm optimization method of neural network weight coefficient for dynamic adjustment, prevent optimization model into local mini- mum, the experiment proved that, this algorithm realization of home appliance control speech recognition accuracy reaches as high as 92%, is the traditional linear system accuracy increased by 20~, and has a strong practical value.

关 键 词：语音识别 优化神经网络 分形维数 粒子群优化 

分 类 号：TP397[自动化与计算机技术—计算机应用技术]