Combining MFCC and Pitch to Enhance the Performance of the Gender Recognition

This paper describes a novel approach which combines the acoustic analysis using MFCC and the speaker´s mean pitch to improve the performance of the gender recognition. In acoustic analysis, two sets of Gaussian mixture model (GMM), male and female, are trained from the speech, and the most likely sequence of models with corresponding likelihood scores are produced. In pitch estimation approach, a threshold is specified to differentiate the two sets. The information provided by the acoustic analysis using MFCC and pitch estimation are combined by using a linear normalization fusion method. The system was tested on the SRMC databases giving at most 3.3% recognition error rate