Voice-Based Detection of Speaker Gender Through Co-Occurrence Matrix-Based Features

Abstract

AbstractGender detection or identification from speech or voice signals is considered a very significant task as it is identified as the preliminary step of speaker recognition. In today’s era, almost all systems are getting converted into smart mode. There are many voice-controlled devices at present, and most interestingly, the number of such devices is getting increased day by day. Speaker gender detection finds its importance in the domain of security also. To provide security in a voice-controlled system detection, the gender of the speaker is very much important. Hence, speaker gender detection is a very much promising research area in the present-day situation. From the performances of voice-controlled devices, it is observed that identification of speaker gets easy if the system first identifies the gender of the speaker as it reduces logical complexity of the system. The gender of the speaker can be identified either through the voice signal of the speaker or through the face image of the speaker. Speaker gender detection through speech signals is comparatively easier than gender detection from facial images. This exertion has considered all the three kinds of gender—male, female as well as trans-gender. Trans-gender has been considered as it is now legally recognized gender. In previous efforts, trans-gender is mostly ignored. The aim of this work is to propose a simple audio facet set to detect all three genders of the speaker. These three genders mostly differ in the frequency domain as well as in the perceptual domain on the point of view of voice signal. For this reason, MFCC-based facets and skewness-based facets have been considered in this exertion. The experimental result proves that the proposed facet set outperforms other previous work.KeywordsGender detectionMFCCCo-occurrence matrixSkewness