基于聚类分析的说话人识别研究

A Modified Pitch Detection Algorithm Based on SIFT

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.