Voice Pathology Detection using Machine Learning and Electroglottography signals

According to a study in the US, 29.9% of the people have a voice disorder during their lifetime while 6.6% have a current voice disorder. Voice pathology detection systems are crucial in areas of disease detection as early detection paves way for the early treatment. A machine learning approach is proposed utilizing Electroglottography (EGG) signals which in turn assesses the vibratory pattern of vocal folds. The developed model uses a binary classifier specifically a Support Vector Machine (SVM) to detect the presence of a voice pathology. Features such as MFCC (Mel Frequency Cepstral Coefficient) Mean, Noise ratio, Spectral Contrast Mean, Open Quotient, Closed Quotient are utilized in the process. These features help assess the vocal behaviour thereby helping the machine learning model analyse patterns with respect to pathological voices. The proposed SVM classifier attained an efficiency of 84% in detecting the voice pathology.