Adaptive noise cancellation and classification of lung sounds under practical environment

Abstract

Lung sound (LS) offers an effective way to detect and discriminate the respiratory disease. However, in practical environments an LS record is subject to serious noise contamination which may be addressed by adaptive noise cancellation (ANC). A least mean square (LMS) algorithm based ANC method is presented by this paper for signal enhancement of LS under practical noisy environment. Based on the hidden Markov model (HMM), minimum classification error (MCE) is adopted to further improve the discriminative performance of LS. Experimental results confirm the effectiveness of the ANC, and the HMM-MCE based lung sounds recognition approach outperforms the traditional HMM-ML(maximum likelihood) method.