Catching the Specific Fusion Features for a Speaker Based on Contourlets

This paper presents the fusion features resulting from contourlet transform on the geometry structure in the time-frequency representation for the one-dimensional speech. In the experiments, after extracting the separate features including main diagonal of the inner lowpass subband (MDL), main anti-diagonal of the inner lowpass subband (MADL), main diagonal of bandpass subband (MDB) and ordinal binary tree difference of main diagonal of bandpass subband(BTDB), the combination features such as MDL-MDB, MDL-MADL-MDB and MDL-MADL-BTDB achieve the remarkable performance with the recognition veracity up to 86.88% enough to match the classical MFCC parameters though the high dimension of the fusion features is still a disadvantage