Cardiopulmonary auscultation is a diagnostic procedure that has a challenging task since the components of heart rate and lung sounds overlap. There were many approaches to quantify the characteristics of these signals, and one of the newest is the voice activity detection (VAD) and the Gaussian Mixture Models (GMM). Considering the lung and heart sounds as acoustic events, this paper proposes a novel assessment methodology of these diagnostic indicators. Here, VAD-GMM was applied to detect and extract the main events in lung sound and heart sounds. VAD-GMM results were compared with other VAD methodology based on statistical approach, and it was found that VAD-GMM give more definite results. Since Mel Frequency Cepstral coefficients (MFCC) and Quartiles feature vectors, were already successful in pattern recognition, VAD-GMM was carried out using this kind of acoustic vectors. Therefore, this method could add in a transition from qualitative traditional auscultation to quantitative assessment and assisted computerized diagnosis by identifying abnormal acoustic indicators. Diagnosis by computerized detection promises to be a more efficient method than traditional methods, which are limited by the auditory capability and experience of a medical professional.
Expanded VAD Guided Subdivision of Cardiopulmonary Sounds
