RespiraCheck: Using Audio Analysis as a COVID-19 Testing Tool

Abstract

To address barriers preventing timely COVID-19 diagnosis, we propose RespiraCheck, a convolutional neural network (CNN) designed to classify COVID-19 based on cough audio. Our approach utilized Mel spectrogram representations of labeled cough recordings to fine-tune the last convolutional and fully connected layers of a pretrained ResNet-18 model, leveraging transfer learning for efficient and accurate classification. Using the crowdsourced Coswara and COUGHVID datasets, we trained on a balanced set of COVID-19 positive and negative samples. To ensure real-world applicability, we also developed a web interface that allows individuals to record or upload cough samples and receive an instant diagnostic assessment. By bridging the gap between clinical research and practical deployment, RespiraCheck aims to provide an accessible, non-invasive, and scalable tool for COVID-19 screening.