ISCA Archive ICSLP 2002
ISCA Archive ICSLP 2002

Data-driven temporal filters obtained via different optimization criteria evaluated on Aurora2 database

Jeih-weih Hung, Lin-shan Lee

In deriving the data-driven temporal filters for speech features, the Linear Discriminant Analysis (LDA) has been shown to be successful in improving the feature robustness [1,2,3]. In our previous works [4,5] it was shown that the criteria of Principal Component Analysis (PCA) and Minimum Classification Error (MCE) can also be used to obtain the data-driven temporal filters in improving the speech recognition performance. In this paper, we proposed to perform Cepstral Normalization before applying these temporal filters, and evaluated the effectiveness of these different data-driven temporal filters on the AURORA2 database. Test results showed very signifi- cant improvements for almost all different cases, specially when the training and testing environments are highly mismatched. Robust MFCC Feature Extraction Algorithm Using