Analysis of rule-based and shallow statistical models for COVID-19 cough detection for a preliminary diagnosis.

Coronavirus pandemic that has spread all over the world, is one of its kind in the recent past, that has mobilized researchers in areas such as (not limited to) pre-screening solutions, contact tracing, vaccine developments, and crowd estimation. Pre-screening using symptoms identification, cough classification, and contact tracing mobile applications gained significant popularity during the initial outbreak of the pandemic. Audio recordings of coughing individuals are one of the sources that can help in the pre-screening of COVID-19 patients. This research focuses on quantitative analysis of covid cough classification using audio recordings of coughing individuals. For analysis, we used three different publicly available datasets i.e., COUGHVID, NoCoCoDa, and a self-collected dataset through a web application. We observed that wet cough has more correlation with covid cough as opposed to dry cough. However, the classification model trained with wet and dry coughs, both, has similar test performance as that of the model trained with wet cough samples only. We conclude that audio-signal recordings of coughing individuals have the potential as a pre-screening test for COVID-19.