Enhancing Wet and Dry Cough Classification with MFCC and Audio Augmentation

This study explores the development and training of a 2-dimensional convolutional neural network (CNN) for the classification of wet and dry coughs using Mel-frequency cepstral coefficients (MFCC) representations. Utilizing a large dataset labeled by four healthcare professionals, known as COUGHVID. This study emphasizes the comparison of different data augmentation techniques: audio data augmentation and MFCC image augmentation. The experimental results demonstrate the efficacy of the CNN model, achieving a testing accuracy of 78%. Furthermore, a comparative analysis between audio and image augmentation techniques provides insights into their impact on classification performance, specifically, a 15% increase in validation accuracy was observed when moving from the original database size to employing both audio and image augmentation techniques. This work contributes to the advancement of cough classification methodologies, with potential applications in healthcare settings for monitoring respiratory conditions and facilitating early detection.