Principal component based classification for text-independent speaker identification

Classification based on Principal Component analysis has recently appeared in the literature in application to text-independent speaker identification. However, results have been reported for only clean speech data. In this paper, we evaluate the performance of principal component classifier for text-independent speaker identification on telephone speech. We then improve its identification performance using a Vector Quantization classifier in combination, through fusion of classifier scores. An identification rate of 78.27% has been obtained on the NTIMIT database, which is well above the best identification rate ever reported in the literature obtained by using only one type of feature set.