An Exploratory Research on Text-Independent Speaker Recognition

This work will handle an exploratory method for text-independent speaker identification. The purpose of the work is to apply a new home-made voice database to the modified algorithm to increase its efficiency for further development and using. Some particular methods of text-independent speaker recognition are described. The proposed techniques do not need signal preprocessing and hence they are time saving easy to implement. The used database obtained from news in a broadcasting channel with varieties of speech tone, emotions, accents in different recording conditions. The feature extraction stage of the proposed approaches is based on Burg's method of linear prediction and the minimal eigenvalues of Toeplitz matrices. The classification part is based on Neural Network - probabilistic and radial types.