MFCC-Based Analysis of Vibratory Anomalies in Parkinson's Disease Detection using Sustained Vowels

Abstract

Parkinson's disease (PD) is a neurological disorder resulting from the degeneration of dopaminergic neurons, which leads to speech production disorders, particularly at the level of the vibratory and articulatory aspects. Early clinical observation of voice disorders can be difficult, hence the growing interest in voice analysis-based diagnoses using machine learning methods. Mel frequency cepstral coefficients (MFCC) are widely used in speech and audio processing as they capture valuable insight about the shape of the vocal tract. The objective of this study is to investigate the relevance of MFCC coefficients as descriptors to identify abnormal vibratory patterns indicative of Parkinson's disease during sustained vowel production. To do this, we evaluate the performance of MFCC in discriminating voices of patients with PD from healthy voices, without combining them with any other speech features. We also perform a comparative study of different classification techniques, such as KNN, SVM, DT, and RF.