Higher Order Spectral Phase Features for Speaker Identification

This paper investigates the use of higher order spectra (HOS) phase features in the task of speaker identification. Within the speech processing community, short time spectral phase information is widely regarded as unimportant for speaker recognition. Features are generally defined from the magnitude spectrum only. This paper utilises features that contain both magnitude and phase spectral information. These HOS phase features are derived by integrating points along a straight line in bifrequency space. Initial experiments used unconstrained, microphone speech from a 20 male speaker database to construct Gaussian mixture models (GMM) for each speaker. The HOS phase features achieve a correct identification rate of 98.5%, which is similar to the rate achieved by the MFCC feature set (99.4%). Other experiments were conducted on the larger YOHO database of 138 speakers. Average correct identification rates of above 95% were achieved for varying populations sizes up to the full 138 speakers.