Unsupervised Feature Selection for Phoneme Sound Classification using Particle Swarm Optimization

This paper proposes a new method based on Particle Swarm Optimization (PSO) for feature selection in phonemes sound classification. Inspired of biologist's studies, each particle is represented by filterbank which is motivated by human hearing. Thus, we propose a technique in which PSO is used to extract audio features similar to human's ear in order to achieve better classification. We use PSO technique for optimizing particle's filterbank in order to classify sound signals accurately. Then, feature extraction is done by using particle's information. Moreover, a classification method based on nearest neighbor is used. Furthermore, by using a defined fitness function in this paper, the particle's features which are represented by a filterbank, can be evaluated and then by using the proposed algorithm, best particle's features can be found. In order to evaluate this proposed method, we create a database which consists of 500 samples for each 12 different phoneme classes. The proposed algorithm is compared with an existing typical audio feature selection based on MFCC. The experimental results shows that the proposed algorithm achieves much better classification accuracy in comparison with MFCC based feature selection method. During iterations, the best fitness value shows remarkable improvement of sound classification accuracy.