Design of Wavelet Based Features for recognition of hindi digits

In this paper WBF (Wavelet Based Features) and LDA (Linear Discriminative Analysis) and PCA (Principal Component Analysis) is used to recognize spoken digits from 0 to 9 in Hindi. The main objective of this work is to find out the wavelet mother that better recognize the spoken Hindi digits. Thirty six experiments are carried out with six different wavelets and for six bands in independent-case. Experiments are done by using six different wavelets Daubechies 10, Daubechies 5, Daubechies 20, Meyer, Coiflet 3, Coiflet 5 and at six different sub-bands 5,6,7,8,9 and 10. Best results have been obtained using wavelets Daubechies 10, and Coiflet 5 at 8 and 9 subbands. The results obtained were compared with MFCC (Mel frequency Cepstral coefficients). Features based on Mel Frequency Cepstral Coefficients (MFCCs) are extracted and their performance is compared with the features extracted by different wavelets. It is found that the recognition performance using Wavelet -based features was superior when compared with MFCC-based features.