基于聚类分析的说话人识别研究  

A Modified Pitch Detection Algorithm Based on SIFT

作　　者：李鹏[1,2] 马道钧[1,2] 余菲[1,2] 

机构地区：[1]西安电子科技大学通信工程学院,西安710071 [2]北京电子科技学院计算机科学与技术系,北京100070

出　　处：《仪器仪表用户》2008年第1期15-16,共2页Instrumentation

摘　　要：通过分析当今说话人识别系统中常用的一些特征参数.提出一种基于改进的K均值聚类算法的多特征组合的说话人识别方法。经过多次实验证明.采用上述方法不仅解决了K均值算法对初始值敏感、易陷入局部最优的问题.而且有效地提高了系统的识别率。Through analyzing some feature parameters that be used usually in speaker recognition system nowadays, a speaker recognition method based on improved K-Means clustering algorithm that combining more features is proposed. Through a series of experiments, it proves that the method not only solves the problems that the K-Means algorithm is sensitive to initial value and lends to local optimizations, but also can improve the recognition rate effectively.

关 键 词：聚类算法 LPCC 基音曲线 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]