Evaluating Speech Production-based Acoustic Features for COVID-19 Classification using Cough Signals

Abstract

Cough is a prevalent acoustic event that contains rich information about underlying ailment in a person. It is used in the diagnosis of several respiratory illnesses including Asthma, Pneumonia, and Tuberculosis. According to medical surveys, cough has been ascertained as a major symptom of the recently declared pandemic, the novel Coronavirus disease (COVID-19). In this work, we attempt to classify COVID-19 positive and negative subjects based on their respective cough recordings. Towards this end, the effectiveness of certain acoustic parameters related to the glottal source and vocal tract of the speech production system, along with spectro-temporal information of the cough signal has been studied for classifying COVID-19 positive and negative samples. These parameters are later used for training a multi-layer-perceptron classifier. The training and performance assessment of this system is done using cough data samples provided in DiCOVA 2021 challenge. Results obtained show that the proposed system outperforms the baseline system, in classifying COVID-19 subjects.