Text independent speaker recognition using the Mel frequency cepstral coefficients and a neural network classifier

Modern speaker recognition applications require high accuracy at low complexity and easy calculation. In this paper, we propose a new method of text independent speaker recognition based on the use of the mean of the Mel frequency cepstral coefficients (MFCC) as a speaker model. These MFCC are extracted from the speaker phonemes in the pre-segmented speech sentences. A multi-layer neural network trained with the back propagation algorithm is proposed to classify these discriminative models. A study is carried out in order to view these models efficiency. Several experiments are made and show that the proposed method gives a high speaker recognition rate. Furthermore, throw these experiments; a technique is proposed to improve this recognition rate by an appropriate phonemes database selection.