Acoustic Features Selection of Speaker Verification Based on Average KL Distance
Yu Luan, Hong Zuo Li, Ya Fei Wang · Applied Mechanics and Materials · 2013
This paper proposes a new Average Kullback-Leibler distance to make an optimal feature selection algorithm for the matching score fusion of speaker verification. The advantage of this novel distance is to overcome the shortcoming of the asymmetry of conventional Kullback-Leibler distance, which can ensure the accuracy and robustness of the computation of the information content between matching scores of two acoustic features. From the experimental results by a variety of fusion schemes, it is found that the matching score fusion between MFCC and residual phase gains most information content. It indicates this scheme can yield an excellent performance.