关键词: 支持向量机, 模糊核子聚类, 非边界聚类区过滤, 聚类区向量集约简

Abstract: Because supporting vectors are near to the border，before training speaker identification system based on SVM using speech training assemble，it is needed to reduce the training assemble.Because the training assemble are tremendous and non-linear，a concept called sub-clusters reduced SVM is presented.Firstly，this paper uses kernel-based fuzzy sub-clusters to filter the non-border cluster areas，then cuts the redundant vectors within the remaining cluster areas.Theory and experimental results show that the time and storage efficiency can be enhanced remarkably by using the training and identifying algorithm，and the system has better robustness.

Key words: Support Vector Machine（SVM）, kernel-based fuzzy sub-clusters, non-border cluster areas filterring, cluster areas vectors reducing