Proposal of Healthcare System for Pathological Symptom Detection Employing Voice Analysis

Mobile healthcare has attracted significant attention in recent years as people have become more health -conscious. Furthermore, there is an increasing demand for non-contact and non-invasive diagnostic methods to reduce the risk of infection and pain. This study proposes a healthcare system that detects pathological symptoms by utilizing speech audio. Existing methods for the detection of COVID-19 have relied on cough recordings. To investigate the potential application of this method in the real use case, this study utilizes speech audio instead of cough recording. Our target is to alleviate the burden on the medical field and help medical workers by implementing this mobile healthcare system. This method uses transfer learning to address the problem of limited data. We evaluate the accuracy with transfer learning and random forest classification utilizing the specific features extracted in such a spectrum and spectrogram.