Hierarchical speaker identification based on PCA and multi-reduced SVM

Abstract

This paper proposed a new hierarchical speaker identification system based on Multi-Reduced Support Vector Machine (MRSVM) and Principal Component Analysis (PCA) classifier to reduce the recognition time of speaker identification. First get a coarse judge by a fast scan of all registered speaker using PCA classifier, and then get a final decision-making by the proposed MRSVM. And the MRSVM has two reduction steps: PCA and kernel-based fuzzy clustering are used to reduce the dimensions and amounts of training data respectively. The experimental results show that the training data, time and storage can be reduced remarkably by using our method, and the system has better robustness.