A filter bank-based source extraction algorithm for heart sound removal in respiratory sounds

In this paper, we consider the problem of heart sounds (HS) removal from respiratory sounds (RS), and a novel semi-blind single-channel source extraction algorithm is proposed. The proposed method is able to extract the underlying pure RS from the HS corrupted noisy input signals by incorporating the filter banks and template-based matching using FIR filters. For performance evaluation of the presented method, the average power spectral densities (PSD) of the input RS segments without HS have been compared with the PSD of the reconstructed signals over six selected frequency bands from 20 to 800 Hz. The proposed method is tested for various types of RS recordings and found effective by yielding an overall maximum spectral difference of 2.8707±0.9875dB for a frequency range below 800 Hz.