Unsupervised detection and classification of heartbeats using the dissimilarity matrix in PCG signals

•An unsupervised approach is proposed to the detection/classification of the heartbeats from phonocardiogram (PCG) signals.•The dissimilarity matrix with the frame-level spectral divergence is a promising tool to estimate heartbeats in PCG signals.•The experimental results have been conducted using several open access databases and a commercial clinical noise database.•The proposed method achieves a significant improvement in heart detection/classification compared to other evaluated methods.•The proposed method shows higher robustness evaluating different types of clinical environmental noises.