Skip to Main Content
In deriving the data-driven temporal filters for speech feature, linear discriminant analysis (LDA) and principal component analysis (PCA) have been shown to be successful in improving the feature robustness. In this paper, we proposed a new data-driven temporal processing method using independent component analysis (ICA) for obtaining a more robust speech representation. ICA is a signal processing technique, which can separate linearly mixed signals into statistically independent signals. The presented method can effectively extract the dominant frequency components ranging between 1 and 16 Hz from the modulation spectrum of speech signals. Detailed comparative analysis between the proposed ICA-derived temporal filters and the previous approaches including LDA and PCA is presented. The preliminary experiments show that the performance of the ICA based temporal filtering is much better in comparison with the LDA and PCA based methods in noisy environment.