Optimization-Based Support Vector Neural network for Speaker Recognition

Speaker recognition is a rapidly emerging area of research. Voice biometrics that is obtained from the speaker's behavior or physical related features provides a pattern of data that accommodate sensitive information about the speaker. The effectiveness of speaker recognition systems is seen to decrease expeditiously due to the mismatch occurrence, including channel degradations and noise. With the aim to assure security and effective recognition, an adaptive fractional bat-based support vector neural network (AFB-based SVNN classifier) is employed to recognize a speaker for which the frequency-dependent features, such as multiple kernel weighted Mel frequency cepstral coefficient (MKMFCC), spectral kurtosis, spectral skewness and autocorrelation are used. The classification is performed using the SVNN classifier based on the extracted features and the classifier is tuned optimally using the proposed Adaptive Fractional Bat algorithm, which is the modification of fractional BAT optimization using the adaptive concept. The experimental result of the proposed method reveals that the accuracy of 0.95, FAR of 0.05, and FRR of 0.05 is obtained, which proved that the proposed method acquired better accuracy and less error value as compared to existing methods.