Comparison of Telephone Recordings and Professional Microphone Recordings for Early Detection of Parkinson’s Disease, Using Mel-Frequency Cepstral Coefficients with Gaussian Mixture Models

Jeancolas, Laetitia; Mangone, Graziella; Corvol, Jean-Christophe; Vidailhet, Marie; Lehéricy, Stéphane; Benkelfat, Badr-Eddine; Benali, Habib; Petrovska-Delacrétaz, Dijana

doi:10.21437/Interspeech.2019-2825

Comparison of Telephone Recordings and Professional Microphone Recordings for Early Detection of Parkinson’s Disease, Using Mel-Frequency Cepstral Coefficients with Gaussian Mixture Models

Laetitia Jeancolas, Graziella Mangone, Jean-Christophe Corvol, Marie Vidailhet, Stéphane Lehéricy, Badr-Eddine Benkelfat, Habib Benali, Dijana Petrovska-Delacrétaz

Vocal impairments are among the earliest symptoms in Parkinson’s Disease (PD). We adapted a method classically used in speech and speaker recognition, based on Mel-Frequency Cepstral Coefficients (MFCC) extraction and Gaussian Mixture Model (GMM) to detect recently diagnosed and pharmacologically treated PD patients. We classified early PD subjects from controls with an accuracy of 83%, using recordings obtained with a professional microphone. More interestingly, we were able to classify PD from controls with an accuracy of 75% based on telephone recordings. As far as we know, this is the first time that audio recordings from telephone network have been used for early PD detection. This is a promising result for a potential future telediagnosis of Parkinson’s disease.