DIAGNOSING PEOPLE INFECTED WITH COVID-19 BASED ON COUGH RECORDINGS

Abstract

Rapid diagnosis of COVID-19 is crucial in preventing the spread of the disease.
There are many solutions such as using early test kits or using new and modern
technologies. One of the solutions, we aim to achieve is the application of artificial
intelligence. There are many other types of research in this area but with the diversity
and rapid growth of new strains, there is no such thing as a complete solution. For
the same reason, we have joined the study to be able to diagnose people with COVID-
19 early through cough recordings. In this paper, we present a CNN model and audio
data augmentation in the COVID-19 diagnosis. The model uses features of MFCCs as
input data. The test results of the models on the Virufy dataset, which are evaluated
based on the ROC AUC results from 85.3% to 98.7%, the AUC of the public test
dataset is 99.5% and the AUC of the private test dataset is 82.9%.