Powered Cepstral Normalization (P-CN) for Robust Features in Speech Recognition

Cepstral normalization has been popularly used as a powerful approach to produce robust features for speech recognition. Good examples of approaches in this family include the well known Cepstral Mean Subtraction (CMS) and Cepstral Mean and Variance Normalization (CMVN), in which either the first or both the first and the second moments of the Mel-frequency Cepstral Coefficients (MFCCs) are normalized. In this paper, an improved approach of Powered Cepstral Normalization (P-CN) is proposed to normalize the MFCC parameters in the r-th powered domain, where r > 1.0. The basic idea is that when the MFCC parameters are raised to the r-th power, the harmful parts of environmental disturbances may be more emphasized than the speech features which are relatively smooth. Therefore performing the normalization in the domain of the r-th power may be more helpful. But the value of r should not be too large because in that case the environmental disturbances may be exaggerated and further corrupt the speech features. This approach is computationally simple and efficient. Initial experimental results on AURORA 2.0 testing environment showed that significant improvements in recognition rates are consistently obtainable under all different noisy conditions.