Gender identification of a speaker using MFCC and GMM

The gender of a speaker, which is the most distinctive characteristics of a speech, can easily be recognized by a person who hears it. Automatic identification of gender information from speech signal is substantially important for many applications. With the identification of gender, gender-dependent systems are defined and accuracy and robustness of systems can be increased. In this study, a system identifying the gender of a speaker independent from a text is developed. The proposed system is based on the classification of MFCC coefficients obtained from speech signals with GMM. In the study, the effect of Gaussian mixture number and MFCC coefficients to the system success is investigated. In the experiments by using TIMIT database, for the 760 sentences of 76 speakers 100 percent success, and for the 6100 sentences of 610 speakers 97.76 percent success is achieved.