Detection of COVID-19 Based on Cough Sound Using LSTM Algorithm

In this study, we propose differentiating positive and negative patients with COVID-19 based on the coughing sound. Cough is an acoustic signal that differs from speech. The difference is in the aperiodic component indicated by the carrier-to-noise ratio. Meanwhile, between positive and negative COVID-19 coughs, non-harmonic components can be analyzed based on a spectrogram, where this feature can use to characterize COVID-19 cough. At the initial stage, experiments carried out the MFCC extracted features in coughing sounds which then entered the classification process in the form of a training and validation process using the LSTM algorithm and obtained an accuracy of 82.2% and a UAR of 71.3% with a training time of 7 minutes 27 seconds. Then after reviewing the imbalance classification data, we used the synthetic minority oversampling technique to synthesize the sample from the minority class in the training dataset before obtaining the classification model. This process can balance the distribution of classes but does not provide any additional information to the model. Then the classification results increased to 91.3% UAR, 90.9% accuracy, with a training time of 1 minute 24 seconds.