Discrimination of male and female voice using occurrence pattern of spectral flux

In this paper, a plain sailing scheme has been propsed for the purpose of discriminating male voice and female voice. Importance of gender identification system is increasing gradually due to its wide application area. Male-female voice discrimination portrays convincing role in the domain of security services, criminal investigation and speaker identification. Voice based identification of male and female is more robust than the other methods currently available. Constant research is going on to classify male and female voice in a better way. Researchers have worked with various types of features including some time domain features like Zero Crossing Rate (ZCR), Short Time Energy (STE) because of its capability to represent the physical characteristics of a certain audio signal. Inspiring from the certitude that male female voice discrimination is well reflected in frequency domain due to the contradiction in biological layout of their vocal cord. Being a frequency domain feature, spectral flux has been adopted for this work. This will expectantly discriminate male and female voice precisely. Spectral-flux based features have been used in the suggested effort which is frequency domain feature. The proposed work deals here to generate a co-occurrence matrix from the spectral flux and then to design feature set based on the extracted features from the matrix. For classification purpose some standard classifiers like RANSAC, k-NN and Neural-Net. The experimental result represents the strength of the proposed feature set.