K-Medoids和FCM融合聚类法语音信号分类的应用

K-Medoids and FCM fusion clustering application research on broadcast and aviation speech signal classification

提出针对广播和航空语音信号的f-kmd融合聚类方法,对2种信号语音数据进行分段,提取每段短时语音数据的均值、方差、平均过零率、平均短时能量、归一化峰度和振幅指标等语音信号的基本特征进行归一化处理,利用模糊c均值聚类(FCM)方法对特征数据进行聚类分析,获得短时分段后的语音信号聚类结果,再对分段后的聚类结果整体上进行K-Medoids聚类分析,得到两类信号的聚类中心。实验表明,融合聚类方法能较好地对广播和航空语音信号进行分类,分类准确率较高,结果较稳定。

A classification method of broadcasting and aviation speech signals is proposed. This method partitions two kinds of voice signal data, and extracts the basic characteristics of each speech signals in a short time, such as the mean value, the variance, the aver- age zero crossing rate, the average short time energy, the normalized kurtosis and amplitude, and so on, and normalizes them. After clus- ter analysis for character data using FCM clustering method, the short-time segmented speech signal clustering results are obtained, and then the results are overall analyzed by K-Medoids cluster analysis, finally the clustering centers of the two kinds of signals are ob- tained. Experiments show that the f-kmd fusion method can preferably classify broadcasting and aviation speech signals with high classi- fication accuracy rate and stable classification results.