An Analysis of Feature Selection based on Optimization Algorithms for Speaker Verification

Speaker verification technology rely on various speaker voice oriented characteristics to discriminate between speakers. Various features like short-term features, temporal features, and long-term features are the importance features used to verify an individual based on the physiological and behavioral characteristics. Various algorithms are available for extracting speaker specific attributes where in recent literature point of view Mel-frequency Cepstral coefficients (MFCC) are most frequently used acoustic features for verifying a speaker accurately based on the characteristics of the vocal track derivatives. Though the verification process is done accurately computation complexity and feature redundancy is the major problem in a speaker verification system. Optimization algorithm is the state-of-art used to reduce feature sets which permit supplementary robust estimates of the model parameters. Also, less computational resources are achieved by using optimal feature selection to overcome time complexity load and redundancy of features during recognition process. In the proposed system Mel-Frequency Cepstral Coefficient (MFCC) are used to extract features and the extracted features are passed on to the optimization algorithm for feature section and finally the computed optimized features are given to GMM classifier for classification. In this article four optimization algorithm like Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Firefly Algorithm (FA) and Grey Wolf Algorithm (GWA) are tested and evaluated for finding the best optimization algorithm for improving the performance of Speaker Verification with reduced feature size and reduced complexity.