Research on Recognition of CHD Heart Sound Using MFCC and LPCC

Abstract

Congenital heart disease (CHD) is a disease that seriously harms the children and family. It needs to be diagnosed and treated in time. The initial diagnosis way of CHD is cardiac auscultation in which the rich experience and expertise are needed. A kind of recognition algorism was proposed to help the initial diagnosis and screening of CHD in this work. The heart sounds were analyzed and the feature extraction by using MFCC and LPCC methods in which the frame length was 2048. The BP neural network was selected as classifier. Results show that the specificity and sensitivity of recognition ratios for CHD are 93.02% and 88.89% by using MFCC, and 86.96% and 86.96% by using LPCC respectively. The features extracted by using MFCC method are better than one by using LPCC.