Feature Extraction and Classification for Automatic Speaker Recognition System – A Review

Automatic speaker recognition (ASR) has found immense applications in the industries like banking, security, forensics etc. for its advantages such as easy implementation, more secure, more user friendly. To have a good recognition rate is a pre-requisite for any ASR system which can be achieved by making an optimal choice among the available techniques for ASR. In this paper, different techniques for the system have been discussed such as MFCC, LPCC, LPC, Wavelet decomposition for feature extraction and VQ, GMM, SVM, DTW, HMM for feature classification. All these techniques are also compared with each other to find out best suitable candidate among them. On the basis of the comparison done, MFCC has upper edge over other techniques for feature extraction as it is more consistent with human hearing. GMM comes out to be the best among classification models due to its good classification accuracy and less memory usage. Keywords— Automatic Speaker Recognition, Mel Frequency Cepstral Coefficients (MFCC), Linear Predictive Cepstral Coefficients, Gaussian Mixture Model ( GMM), Vector Quantization ( VQ), Dynamic Time Warping (DTW), Hidden Markov Model ( HMM), Wavelet decomposition