“…This capability of deep networks makes them the right tool with which to extract and learn particular patterns from diverse and relatively large training sets, such as heart sound databases (PhysioNet, 2016). However most existing AI-assisted heart sound analysis systems use medically incomprehensible mathematical and statistical features, such as wavelets (Clifford, 2016), Stockwell transformation (Moukadem, Dieterlen, & Brandt, 2013), Mel-frequency cepstral coefficients (MFCCs) (Chen, 2017), and the likelihood function (Yamashita, Himeshima, & Matsunaga, 2014).…”