PERFORMANCE EVALUATION OF SPEAKER IDENTIFICATION SYSTEM FOR MALE SPEAKER DURING ADOLESCENCE

A voice mutation commonly refers to the deepening of voice of people from early to late adolescent stages. Males experience a more dramatic voice mutation than females during adolescence. This paper presents experiment for performance evaluation of speaker identification system for male speaker during adolescent stages. A text independent, closed set speaker identification system is implemented using MelFrequency Cepstral Coefficients (MFCC) and Dynamic Time warping (DTW) algorithms in MATLAB for training and testing of recorded short sample sentences of native Hindi male speaker during adolescent stages for speaker identification accuracy. The system achieved approx 98% speaker identification accuracy for the speech sentences trained against sample sentence of corresponding age but speaker identification performance was insignificant when trained speech sample sentence was of different age from the age of testing speech sentences. The results show strong influence of voice mutation on the speaker identification and the voice biometrics during adolescent stages may need careful sampling of speech data.