Study of automatic biosounds detection and classification using SVM and GMM

Ambulatory devices can be used to detect heart diseases and save lives in critical time. These devices are based on sound classification that usually adopts a suitable data mining algorithm. This paper investigates the performance of Support Vector Machine (SVM) and Gaussian Mixture Model (GMM) classifiers in classifying sound samples. SVM classifier makes use of a linearly separable hyperplane to classify data into different classes, while GMM utilizes a probabilistic model for density estimation through probability density functions. Feature vectors of sound samples were extracted using the Mel-frequency cepstral coefficients (MFCCs) and fed to the classifiers. Our experimental results showed that SVM is more robust than GMM, and SVM achieved >80% classification accuracy in all classes of sound samples collected in this study.