On the pragmatism of using binary classifiers over data intensive neural network classifiers for detection of COVID-19 from voice