A Neoteric Speaker Verification System Using Ant Colony Optimization Algorithm

Automatic Speaker Verification (ASV) systems are mostly being used for authentication of a known identity based on different voice features. The ASV involves a comparison of a speaker's voice with pre-stored training data or speaker model captured during the enrolment phase and also with an impersonator model of non-identical voices differing in gender as well as phone types. Thereafter, a score is assigned by the system and decides whether the test speaker is an enrolled speaker or an imposter. In the comprehensive classification of a certain problem, one of the most important processes is feature subset selection. It includes dimensionality reduction, attribute subset selection and variable subset selection and also an important step to reduce overfitting of data. This paper presents one of the methods to solve this feature selection problem using Ant Colony Optimization algorithm. This algorithm subsumes the performance of the classification into state transition rule. The optimal feature set is determined with the help of the pheromone from the ant colony. Finally, a feature set with maximum classification accuracy and small size is selected.