Application of a Hybrid Filter for Audio Features Selection in Speaker Recognition Systems

Speaker recognition systems have high dimensional audio features which most of them are irrelevant and redundant. So far, many approaches have been proposed to reduce the size of these features which most of them are based on the system training and evaluating the performance of each feature subset. In this paper, a hybrid semi-wrapper method is proposed to select the audio features in speaker recognition systems. The objective of this algorithm is to increase the accuracy of the speaker recognition system using supported vector machine (SVM) as a classifier. In the first step, genetic and ant colony algorithms were used to select the optimal subset of features. Then, the proposed technique was presented for selecting spectral features. SVM classifier was used to achieve better accuracy in recognition of real speaker. This algorithm was evaluated based on the error rate. The results showed that the number of selected features, time complexity and error rate of the proposed method are less than other methods.