Exploring Artificial Intelligence in COVID-19 Detection Through Cough Sound Analysis

Previous studies feature cough sound recordings and AI for COVID-19 pre-screening. These studies often neglect cough segmentation and relies on single-source datasets. This study introduces a deep learning framework with cough segmentation methods using 15 MFCC coefficients and a Mini VGGNet model. Cross-dataset training and testing with Cambridge, Coswara, and NIH Malaysia datasets were also conducted. Findings highlight manual segmentation (0.4-second duration, 0.1-second overlap) as the optimal cough segmentation method. The best performing model was trained on 80% of merged dataset (triple sourced) and tested on 20% of the Cambridge set achieving 0.921 test-accuracy, 0.973 AUC, 0.910 precision, and 0.910 recall.