基于MFCC和LPCC的说话人识别  

Speaker recognition method using MFCC and LPCC features

作　　者：余建潮[1] 张瑞林[1] 

机构地区：[1]浙江理工大学,浙江杭州310018

出　　处：《计算机工程与设计》2009年第5期1189-1191,共3页Computer Engineering and Design

基　　金：浙江省自然科学基金项目(Y106148)。

摘　　要：MFCC参数和LPCC参数是说话人识别中两种最常用的特征参数,研究了MFCC和LPCC参数提取的算法原理及差分倒谱参数的提取方法,采用MFCC、LPCC及其一阶、二阶差分作为特征参数,通过k均值算法与三层BP神经网络来进行说话人识别。实验结果表明,该方法可以有效提高识别率,同时也验证MFCC参数的鲁棒性优于LPCC参数。MFCC and LPCC are the two most important feature parameters in speaker recognition. On the algorithm principle of MFCC and LPCC parameter extraction and the extraction method of coefficient cepstral parameter are researched. The Speaker recognition is realized by K-means algorithm and three-layer neutral network, adopting MFCC, LPCC and accelerated coefficient as feature parameter. The result shown that this method can efficiently accelerate the recognition capacity of the system, meanwhile, it proves that the robustness of MFCC parameter is prior to that of LPCC parameter.

关 键 词：LPCC MFCC 特征提取 说话人识别 神经网络 K均值算法 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]