Optimal MFCC features extraction by differential evolution algorithm for speaker recognition

Speech is the most commonly and widely used form of communication and interaction between humans. The interfacing system, which is an automatic speaker recognition system, requires modeling to receive input data in the form of a feature with a minimum number and learn through this data. The purpose of this paper is to extract the optimal number of Mel-Frequency Cepstral Coefficients (MFCC) features without reducing the recognition accuracy for speaker recognition application. For this purpose, an algorithm has been proposed in which the Differential Evolution (EA) optimizer and also the probabilistic neural network (PNN) classifier are used to achieve this goal. After implementing this algorithm in MATLAB software, it was observed that the number of MFCC features, which so far had at least 13 for each frame, was reduced to 5 per frame, without any recognition accuracy being reduced.