Exploring Potential Audio Features for Respiratory Disorder Detection with COVID-19

The COVID-19 pandemic has presented unprecedented challenges to public health, with the detection and isolation of potentially infected patients being one of the most crucial aspects of pandemic control. Physical testing for COVID-19 can be difficult due to the fear of infection. In this study, we have proposed a model that utilizes cough audio features to detect COVID-19. Deep learning models are used to extract acoustic parameters from cough data-sets, which, when combined, provide a more comprehensive representation of the audio signal and help to distinguish between different classes of audio signals. The extracted features are then used to create spectrograms and four different deep learning models (CRNN, ResNet, Inceptionv3, and VGG16) to classify cough audio. The models demonstrate fair performance in classifying cough noises, with CRNN being the most successful among the others. Our approach shows potential for further development into a preliminary COVID-19 detection application, which may be applied to identify other respiratory disorders such as pneumonia, asthma, tuberculosis. The deep learning models, together with the acoustic properties obtained from cough noises, provide a promising route for further study in the area of respiratory disease diagnosis. Our findings suggest that cough and breathing audio features can be utilized for detecting of COVID-19 and other respiratory disorders in the future.