Respiratory disorder classification based on lung auscultation using MFCC, Mel Spectrogram and Chroma STFT

A significant portion of the population suffers from various lung function disorders on a daily basis, which ultimately result in respiratory problems. For respiratory disorders to be managed effectively, prevention and early identification are crucial. Lung sound analysis has attracted more attention recently. So it’s likely that this discipline might one day allow for the automated inference of irregularities prior to respiratory collapse. An effective predictive model is required to reduce fatalities. The paper contrasts several feature extraction techniques applied in respiratory disorder classification models and offers an integrated solution for the issue. In this work, lung auscultation recordings are used to train a two-dimensional convolutional neural network (CNN) to identify respiratory diseases. In comparison to other models, the integrated solution significantly reduced the loss and attained an accuracy of 94.90%.