Multi-layered Features with SVM for Text-independent Speaker Verification

In this paper, we propose an approach for text-independent speaker verification system usage both cepstral and prosodic features. We combine MFCC features and pitch contour features to capture the characteristics in speaker verification. We are using cubic polynomials to estimate the pitch contour segments in order to model the differences on intonation contour, and using the support vector machine (SVM) to measure overall system´s effectiveness. Experimental results show that the proposed approach can significantly improve the text-independent speaker recognition.