Publication

Abstract : Brain diseases, which encompass a wide range of conditions and illnesses brought on by stroke, Alzheimer’s, Parkinson’s, traumatic brain injury, and many other conditions, afflict 1 in 6 individuals worldwide. One such disorder that gradually destroys nerve cells in the midbrain is Parkinson’s disease. It is a neurodegenerative condition that impairs motor abilities. Additionally, it has an impact on the muscles involved in speech production, leading to hypokinetic dysarthria, a set of motor speech disorders that includes dysphonia, bradylalia, and poor articulation accuracy. In this context, this work proposes the application of recurrence plots to map the speech signals onto images, which will be further fed into a Convolutional Neural Network for the automatic classification of PD from healthy controls. The proposed approach is assessed on the Italian Voice and Speech Data, containing Diadochokinetic (DDK) recordings of /pa/ & /ta/ audio recordings. The experimental results of the full audio file approach produced an average testing accuracy of 83%. Also, the model based on the frame-based approach performed well on the test set, resulting in an average test accuracy of 91% for both /pa/ and /ta/ recordings.