Speaker identification based on a novel TDS

Abstract

SVM has been used in speaker identification successfully, whereas training SVM consumes long computing time and large memory with all training data, therefore the training data selection (TDS) is an important step for effective speaker identification system. In this paper, a novel TDS method based on the PCA and improved ant colony cluster (IACC) is proposed to solve this problem existed in SVM. The proposed TDS method has two steps. Firstly, the PCA-based feature selection approach is exploited to reduce the dimension of the input vectors, and then the IACC is used to select the center data of each cluster to reduced training data. Experimental results show the training data and the storage can be reduced greatly, and the proposed system has better identification performance and robustness than other model.