Recognizing similar acoustic sounds using wavelet packets

This paper investigates the performance of the Wavelet Packet Scale (WPS) in the analysis of manually generated subwords. This scale was derived by the authors in previous work [9]. The set of words chosen is the A-set which is a subset of the alphabets. It contains the letters a, j and k. The parameterization of the subwords is accomplished using energy coefficients following the WPS model. Radial Basis Function Artificial Neural Network (RBF-ANN) is employed for the recognition tasks. We investigated the success of the proposed model with two orthogonal mother wavelets from the Daubechies set and one spline biorthogonal wavelet. The performance of these wavelets was compared with the traditional Mel scale.