Gammatone Visualization based Cough Sound Classification: Performance Comparison with Delta and Delta-Delta Features

Abstract

Cough being a common symptom for most respiratory disease is considered as a predictor in the diagnosis of the diseases. In recent years, time frequency representations of signals are acclaimed for its efficacy in the classification of signals. This work explores the potential of time frequency representation derived from gammatone features in the classification of cough signals. Accordingly, visualization of gammatone cepstral coefficients (GTCC) and its delta and delta-delta variants are employed for classifying cough signals using GoogLeNet, a prominent pre-trained CNN architecture. The results of this study evinces that the delta-delta variant of GTCC with a classification accuracy of 98.02% has significantly outperformed GTCC and its delta variant which recorded accuracies of 97.22% and 94.44% respectively.