COVID-19 Detection Model with Acoustic Features from Cough Sound and Its Application

Contrary to expectations that the coronavirus pandemic would terminate quickly, number of people infected with virus did not decrease worldwide and coronavirus-related deaths continue occur every day. The standard COVID-19 diagnostic test technique used today, PCR testing, requires professional staff equipment, which is expensive takes a long time produce results. In this paper, we propose feature set consisting four features: MFCC, Δ2-MFCC, Δ-MFCC, spectral contrast as optimized for diagnosis COVID-19, apply it model combines ResNet-50 DNN. Crowdsourcing datasets from Cambridge, Coswara, COUGHVID are cough sound data our study. Through direct listening inspection dataset, audio recordings contained only sounds were collected training. was trained tested using features extracted crowdsourced had sensitivity specificity 0.95 0.96, respectively.