A Cepstral PDF Normalization Method for Noise Robust Speech Recognition

Abstract

In this paper, we propose a novel cepstrum normalization method based on the scoring procedure of order statistics for speech recognition in additive noise environments. The conventional methods normalize the mean and/or variance of the cepstrum, which results in an incomplete normalization of the probability density function (PDF). The proposed method fully normalizes the PDF of the cepstrum, providing an identical PDF between clean and noisy cepstrum. For the target PDF, the generalized Gaussian distribution is selected to consider various densities. In recognition phase, a table lookup method is devised in order to save computational costs. From the speaker-independent isolated-word recognition experiments, we show that the proposed method gives improved performance compared with that of the conventional methods, especially in heavy noise environments.