Abstract:As heart sound signal in strong noise levels was denoised using traditional heart sound denoising algorithm, it is easy for part of heart sound to be reguarded as noise and removed, which leads to the energy loss of available heart sound. A multi-stage singular value decomposition (MS-SVD) algorithm was proposed and used to pick up the principal components of heart sound based on the feature of the principal components analyzing of singular spectrum analysis. In order to remove the noise in low frequency, the heart sound extracted was decomposed into low-frequency coefficient and high-frequency’s with wavelet packet, and using adaptive threshold method dealed with the low-frequency coefficient. The high-frequency’s was also analyzed to pick up the heart sound in it according to the multiresolution characteristics of wavelet packet. The simulation results illustrated that it can improve obviously the indexes of SNR, SNRG and RMSE, and the denoising performance in different noise levels is superior to the traditional method. The experiments also demonstrated that it not only can remove the noise components in heart sound efficiently, and also reserve the details of the original signal.