A Study of Feature Parameters Based on LPC Analysis with Applications to Speaker Identification
Zhen Yang · Journal of Nanjing University of Posts and Telecommunications · 2005
In this paper,LPC predictor coefficients and LPC-derived coefficients are studied and compared from the point of view of computation method.A new set of features composed of LPC coefficients and speech frame energy is introduced.Closed-set text-independent speaker identification experiments with 20 speakers are conducted using a GMM classifier.The experimental results show that,compared to LPC-derived coefficients,the proposed feature parameters can provide comparable accuracy with lower computational complexity and compared to LPC coefficients,the proposed feature parameters can yield significantly improved performance with slightly higher computational complexity,especially for short test utterance.