Fig. 7.
COVID-19 speech classification: An SVM classifier using bottleneck features (BNF) achieved the highest AUC of 0.923 when classifying COVID-19 speech in ComParE dataset. A Resnet50 trained by transfer learning achieves a slightly lower AUC of 0.914. Speech (normal and fast) in the Coswara dataset can be used to classify COVID-19 with AUCs of 0.893 and 0.861 respectively using a Resnet50 trained by transfer learning.