The Use of Perceptual Energy in Speaker Discrimination

The short-term energy is widely used in various speaker discriminating applications (such as speaker verification system, diarization system and etc). However, unlike the well-known loundness measurement, its calculation takes no considerations on the human auditory properties. Since it has been shown that PLP is, to some extent, better than MFCC due to its well approximation of the excitation pattern to the human hearing system, the perceptual energy is thus introduced in this paper to replace the short-term energy, which is calculated to the PLP framework. It has been demonstrated that the newly introduced energy measurement exhibits more disseminating power with less computational overhead than the classical short-tem loudness. Its effectiveness is further verified in the application of speaker diarization.