This article has been accepted at the IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2024
Acoustic signals are crucial for health monitoring, particularly heart sounds that provide essential data such as heart rate and detect cardiac abnormalities such as murmurs. This study uses a publicly available phonocardiogram (PCG) dataset to estimate heart rate using model-based methods and extends the best-performing model to a multi-task learning (MTL) framework for simultaneous heart rate estimation and murmur detection. Heart rate estimates are derived using a sliding window technique on fragments of heart sounds, analyzed with a combination of acoustic features (Mel spectrogram, cepstral coefficients, power spectral density, mean square energy). Our findings indicate that a 2D convolutional neural network (2dCNN) is most effective for heart rate estimation, achieving a mean absolute error (MAE) of 1.312 bpm. We systematically investigate the impact of different feature combinations and find that using all four features yields the best results. The MTL model (2dCNN-MTL) achieves over 95% accuracy in murmur detection, outperforming existing models, while maintaining an MAE of 1,636 bpm in heart rate estimation, meeting the requirements set by the Association for the Advancement of Medical Instrumentation (AAMI).