A Gaussian mixture models approach to human heart signal verification using different feature extraction algorithms

In this paper the possibility of using the human heart signal feature for human verification is investigated. The presented approach consists of two different robust feature extraction algorithms with a specified configuration in conjunction with Gaussian mixture modeling. The similarity of two samples is estimated by measuring the difference between their negative log-likelihood of the features. To evaluate the performance and the uniqueness of the presented approach tests using a high resolution auscultation digital stethoscope are done for nearly 80 heart sound samples. The experimental results obtained show that the accuracy offered by the employed Gaussian mixture modeling reach up to 100% for 7 samples using the first feature extraction algorithm and 6 samples using the second feature extraction algorithm and varies with average 85%.