Feature Selection using Singular Value Decomposition for Stop Consonant Classification

In the research field of pattern recognition, especially in the signal classification, the process of determining the suitable and the relevant feature is important to obtain the better classification result. This paper presents the feature selection of stop consonant by using singular value decomposition (SVD) and the classification by applying multi-layer perceptron (MLP). The feature sets were derived by using the wavelet packet transform (WPT) at 4th decomposition level with daubechies2 wavelet family, and the WPT-based feature after being dimensionality reduced by using SVD with varying reduction index which denotes as SVD1 and SVD2. Each CVC stop consonant is windowed at a certain length of duration to obtain a relevant CV unit. The experimental result shows that SVD gives improved classification scores.