DISCRIMINATIVE FEATURE EXTRACTION BASED ON SELF-ADAPTIVE FREQUENCY WARPING FOR ROBUST SPEAKER IDENTIFICATION

摘要

This paper presents a new discriminative feature based on self-adaptive frequency warping. We analyze the discrimination power between frequency components and individual characteristics and quantify this dependency. This new feature is extracted by nonuni-form sub-band filters designed according to self-adaptive frequency warping in different frequency bands. Furthermore, in order to overcoming the acoustics mismatch between training and testing data in the noise environment, we adopted pre-enhancement prior to feature extracted module. Using a series of controlled experiments, it is shown that the theory of this feature is reasonable and understandable, which is insensitive to spoken content and thus more discriminative and robust in comparison to the conventional Mel frequency cepstral coefficients. The experimental results demonstrate that combining pre-enhancement and discriminative feature leads to noticeable improvement on speaker recognition rate and robustness.