Deep CNN for Voice Pathology Classification using Electroglottography

The increasing frequency of voice disorders highlights how crucial early diagnosis is to successful treatment and intervention. By using a novel approach, this work addresses the pressing need for speech pathology detection. Electroglottography (EGG) signals from the SVD (Saarbruecken Voice Database) database is utilized in our research to distinguish between the samples from healthy and pathological conditions. The MFCC (Mel Frequency Cepstral Coefficients) serves as discriminative feature input to the Convolutional Neural Network (CNN) classifier. This research has worked on implementing the identification of 12 various voice disorders such as Dysodie, Cyste, Dysarthrophonie, etc. from the SVD dataset. The implemented CNN classifier has attained a remarkable accuracy of around 94% in the detection of disorders.