Phonocardiogram (PCG) signal is a graphical representation of the heart sounds that can be used to diagnose a heart disease. Diagnosing heart disease based on PCG signal is more effective. Because of its ability to capture all heart sound components including S1 and S2. Nevertheless, the interpretation of PCG signal is depend on the cardiologist's expertise. Therefore automated PCG signal classification is required in order to help the cardiologist diagnosing and monitoring heart disease. The classification of PCG signal is influenced by the segmentation and the feature extraction process. The segmentation process aims to detect the location of heart sound components including S1 and S2 in PCG signal. However it is difficult to find those component in a noisy PCG signal. The feature extraction process aims to extract relevant features that lie in segmented PCG signal. This process is required because the segmented PCG signal has high dimensionality and redundant information. This study proposes Shannon Energy Envelope for segmenting PCG signal and Deep Belief Network (DBN) for feature extraction method. The results show that the proposed method outperforms shallow models in existing datasets.