Optimization of filter-bank to improve the extraction of MFCC features in speech recognition

Mel-frequency cepstral coefficients (MFCC) have been demonstrated to perform very well under most conditions. However, some limited effort has been made to optimize the shape of the filters in the filter-bank using the conventional MFCC approach. This work develops several new approaches to designing the shapes of filters in the filter-bank. In these new approaches, principal component analysis (PCA) and linear discriminant analysis (LDA) are modified and then used to generate new filters. The experimental results reveal that the proposed approaches can improve the recognition performance of MFCC in noisy environments.