Speaker Identification Using HHT Spectrum Features

This paper proposes new acoustical features based on Hilbert Huang transform (HHT) for speaker identification. HHT is a powerful analysis method to obtain instantaneous frequency (IF). First, empirical ensemble empirical mode decomposition (EEMD) is used to generate intrinsic mode functions (IMFs). The Hilbert transform is then applied to IMFs to compute the instantaneous frequencies. With the obtained instantaneous frequencies, two new acoustical features are presented. The first acoustical feature is the weighted mean IF in each IMF while the second is the IF difference between two consecutive IMFs. This study adopts Gaussian mixture model (GMM) to train and test the speaker models. Finally, the experiments conducted on CHAIN corpus demonstrate the superiority of the proposed acoustical features.