A Novel Gender Recognition in Emotional Environment using GMM

The emotional speech is considered as a prominent aspect in human speech communication. Emotional Speech based gender recognition has its efficacy in real world applications like biometrics, human computer related fields, voice synthesis etc. The goal of gender recognition is to extract the information from speech signal depending on speaker identity. The recognition of the gender of the speaker in emotional state is main premise of this proposed work. Both prosodic features and spectral domain features are used. The parameters of prosodic features include pitch, energy and formant; spatial features include Mel Frequency Cepstral Co efficient (MFCC). Gaussian Mixture Model is used as classifier to train and classify the gender from the features extracted from Berlin database. The accuracy of the proposed techniques is measured using metrics like Sensitivity, Specificity and likelihood ratio.