Selective use of the speech spectrum and a VQGMM method for speaker identification

Abstract

This paper describes two separate sets of speaker identication experiments. In the rst set of experiments, the speech spectrum is selectively used for speaker identication. The results show that the higher portion of the speech spectrum contains more reliable idiosyncratic information on speakers than does the lower portion of equal bandwidth. In the second set of experiments, a v ector-quantization based Gaussian mixture models (VQGMMs) is developed for text-independent speaker identication. The system has been evaluated in the recent speaker identication evaluation organized by NIST. In this paper, details of the system design are given and the evaluation results are presented.