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

This paper describes two separate sets of speaker identi cation experiments. In the rst set of experiments, the speech spectrum is selectively used for speaker identi cation. 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 vector-quantization based Gaussian mixture models (VQGMMs) is developed for text-independent speaker identi cation. The system has been evaluated in the recent speaker identi cation evaluation organized by NIST. In this paper, details of the system design are given and the evaluation results are presented.