Principal factor analysis and SVM based effective speaker recognition

Abstract

Speaker recognition is important for successful development of speech recognizers in various real world applications. In this paper, the speaker recognizer was developed using sizable collection of various speakers both male as well as female with pitch strength as the feature. We proposed Principal Factor Analysis (PFA) technique for dimensionality reduction for accurate speaker recognition system. The first module performs feature extraction from speech samples taking pitch strength as the feature. The second module executes dime-nsionality reduction from the windowing of speech samples, where data samples are normally signified as matrices or higher order tensors. The system was trained by Support Vector Machine (SVM) using dimensionality reduced feature matrix. The implementation results show that the proposed system recognizes whether the given speaker is authorized or not.