Feature Extraction Based on Wavelet Transformation in Speaker Recognition
Liu Ya · Journal of Luoyang Institute of Technology · 2005
In speaker recognition system, one of the key problems is to extract the valid speech features that can represent speaker's characters.According to a research in wavelet theory and derived from a conventional feature parameters, MFCC,which is based on human auditory mechanism,a new feature parameters have been presented.The experiment results indicate that the new feature parameter DWTMFC is better than MFCC.After comparing and discussing several wavelet functions in common use,a preferable wavelet function (coif3) is selected.