Impulse noise reduction in audio signal through multi-stage technique

Abstract

The fundamental aim of this study is exclusively focused upon detecting and restoring audio signals corrupted by random-valued impulse noise. This study is stemmed from a process in which the noisy samples are in depth analysed through a cascade of stages. In each stage, a vector of different length is used and the similarity degree between the tested sample and its neighbors are precisely measured in each vector. The tested sample which achieves the minimum degree of similarity with its neighboring samples in each stage is highly regarded as an original sample. Several experiments are carried out on different audio sounds to assess the performance of the proposed algorithm in terms of peak signal to noise ratio (PSNR) and sound quality. The attained results have shown that the new method provides remarkable performance if compared to other well-known algorithm, particularly at low noise rates. The method adopted has turned out to be more successful in removing noisy samples and keeping the original ones intact. Also the method in question highlights low computational complexity and easy to implement.