Feature extraction of heart sounds using velocity and acceleration of MFCCs based on support vector machines

Heart sounds Auscultation has usually used as an important primary symptom to identify cardiovascular diseases (CVDs). In this paper, a new approach of automatic recognition of normal and abnormal heart sounds (HSs) is produced. Features of HSs are extracted by several steps. First, signals of heart sounds are preprocessed. Second, Energies of these signals are calculated and divided by logarithmic of these energies. Then, velocity and acceleration of MFCCs (mel frequency cepstral coefficients) are conducted for the second step. Fourth, Statistical calculations are calculated for the third step to get the features of heart sounds. Support vector machines (SVMs) are used as classifiers for the extracted signals. The obtained recognition percent is 98.44%.