The Automatic Repairing Method Addressing Clipping Distortions and Frictional Noises in Electronic Stethoscope

Abstract

The auscultation signal collected by the electronic stethoscope may be sometimes accompanied by various interferences, including external speech/acoustic interferences, clipping distortions, frictional noises, etc. The external speech/acoustic interferences can be eliminated by adaptive filtering, with the aid of an extra recording sensor. However, clipping distortions and frictional noises cannot be addressed by this methodology, and how to automatically repair them has not been fully discussed in the literatures, which affects the signal quality and further the cardiopulmonary sound automatic diagnosis. In this paper, the repairing method that automatically addresses clipping distortions and frictional noises for electronic stethoscope is developed. A simple signal difference method is introduced to automatically detect the clipping distortion regions, and these regions are repaired by the Hermite interpolation. The regions that frictional noises exist are detected by employing Mel-frequency cepstral coefficients (MFCCs) and support vector machine (SVM), and they are repaired by involving the empirical mode decomposition (EMD) as well as correlation coefficients. The proposed method can automatically detect, locate and ultimately repair multiple regions of clipping distortions and frictional noises, and applying it in recorded real auscultation data proves its efficiency.