基于减法聚类与改进的模糊C-均值聚类算法的说话人识别方法的研究

Research on Speaker Recognition Method Based on Subtractive Clustering and Improved Fuzzy C-means Clustering Algorithm

提出一种将减法聚类与改进的模糊C-均值聚类相结合并用于说话人识别的方法.该方法将从语音信号中提取的Mel频率倒谱系数及其差分作为特征参数;用减法聚类算法初始化聚类中心,再用改进的模糊C-均值聚类算法进行修正,形成码本.识别时,对每一个待识别语音进行模糊聚类识别.仿真结果表明,该方法比改进的模糊C-均值聚类算法识别率高,具有较好的鲁棒性,且计算比较简单.

A speaker recognition method based on subtractive clustering and improved FCM （Fuzzy C-Means） clustering is introduced, and the method uses Mel frequency cepstrum coefficient （MFCC） and its difference which are extracted from speech signals as the characteristic parameters. Subtractive clustering is utilized to initialize the cluster centers, improved FCM clustering is utilized to revise the cluster centers, and then the code book is formed. In recognition, each unrecognized speech signal is clustered to the code book through FCM. The simulation results reveal that the presented method has a higher recognition rate, a better robustness and a simpler computation than those of the improved FCM.