Histogram Classifiers Using Vocal Tract and Pitch Information for Text-Independent Speaker Identification

This paper proposes a new textindependent speaker identification method based on histogram classifiers using vocal tract and pitch information. The purpose of this research is to create a very simple and fast speaker identification method. The method uses several feature parameters (cepstral coefficients, pitch) related to both vocal tract and pitch information extracted from spoken sentences. This paper shows that the pitch is an additional important characteristic which improves significantly the speaker identification performances. This proposed method, evaluated using a subset of the DARPA TIMIT database composed of 462 speakers, provides good results which encourage further research. I. Introduction At present, most of the speaker recognition methods (DTW,HMMs,NN,AR) are based on temporal correlation. The method proposed in this paper is independent of this aspect and still provides good results. The latest research on the Auto-Regressive (AR) Vector Model [1] tends to show that ..