基于人工小波神经网的说话人识别  

Speaker Identification Based on Wavelet-artificial Neural Network

作　　者：赵孔新[1] 丁宁[1] 

机构地区：[1]长春工业大学计算机科学与工程学院,吉林长春130012

出　　处：《电声技术》2008年第7期51-53,共3页Audio Engineering

摘　　要：结合小波变换时频分析的局部化特性和人工神经网的良好学习能力,设计了一种用于声纹识别的人工小波神经网,建立了基于显式小波的网络学习算法。计算机模拟表明,与传统的算法相比,用该小波神经网提高了语音信号分类识别的速度和识别率,而且结构简单易于添加语音模板以便增加识别说话人的人数,是一种可行的语音识别系统。The wavelet-artificial neural network is used for speaker identification, which unites the localization characteristic of wavelet transform and unique study ability on artificial neural network. The learning process is based on the algorithm of the explicit wavelet. The results of the simulation indicates that comparing with the traditional algorithm, this kind of wavelet artificial neural network has the faster speed and the better identification rate, moreover its structure is simple and adaptable for needs of adding more speakers.

关 键 词：人工小波神经网 模式识别 拓扑结构 训练算法 

分 类 号：TN912[电子电信—通信与信息系统]