通过核 K- 均值聚类的方法对语音帧进行聚类 , 由于聚类的中心能够很好地代表类内的特征, 用中心样本帧取代该类, 减少了核矩阵的维数, 然后再采用稀疏 KPCA方法对核矩阵进行特征提取。-Through the nuclear K-means clustering method for clustering of speech fr a mes, the cluster center can be a good representative of the class characteristics of the sample fr a me to replace the class with the center, reducing the dimension of the nuclear matrix, and then use Sparse KPCA method for feature extraction of the nuclear matrix.