Robust speaker identification using vocal source information

Abstract

This paper highlights the effectiveness of Wavelet Octave COefficients of Residues (WOCOR) based feature extraction for robust text-independent speaker identification. A new feature set, WOCOR is proposed to capture the spectro temporal source excitation characteristics of the speech signal. This work is focused to increase the identification accuracy with databases containing short length speech signal. Experimental evaluation is carried out on TIMIT database with 630 speakers using Gaussian Mixture Model (GMM) is used as classifier. Vocal source feature is used to extract the information from the residual signal. The vocal source information contains pitch, pitch frequency and phase in the residual signal. In this project, 93.02% Identification rate is achieved in WOCOR.