Cough Classification for COVID-19 based on audio mfcc features using Convolutional Neural Networks
Vipin Bansal, Gaurav Pahwa, Nirmal Kannan
Abstract
Cough analysis is an important area in the clinical diagnosis of various respiratory illnesses. Cough is the body's inundate mechanism to clear the throat and lungs of foreign material. It presents a great opportunity for the initial assessment of potential patients infected by COVID-19 which is a new virus. With symptoms like other respiratory illnesses, dry cough, fever and fatigue, it is challenging to classify probable Covid infection with no Covid data available yet. However, we propose a CNN based audio classifier [1] using the open cough dataset. The dataset is labeled manually into cough categories with final labeling into Covid and Non-Covid classes. The two approaches proposed in this paper are based on mfcc features and spectrogram images as input to CNN network. MFCC approach produced 70.58% test accuracy with 81% sensitivity and is better than the spectrogram-based approach.