The Gaussian Mixture Model Universal Background Model (GMM-UBM) approach is used to assess the Parkinson’s disease (PD) progression per speaker. The disease progression is assessed individually per patient following a user modeling-approach. Voiced and unvoiced segments are extracted and grouped separately to train the models. Additionally, the Bhattacharyya distance is used to estimate the difference between the UBM and the user model. Speech recordings from 62 PD patients (34 male and 28 female) were captured from 2012 to 2015 in four recording sessions. The validation of the models is performed with recordings of 7 patients. All of the patients were diagnosed by a neurologist expert according to the MDS-UPDRS-III scale. The features used to model the speech of the patients are validated by doing a regression based on a Support Vector Regressor (SVR). According to the results, it is possible to track the disease progression with a Pearson’s correlation of up to 0.60 with respect to the MDS-UPDRS-III labels.
Cite as: Arias-Vergara, T., Vasquez-Correa, J.C., Orozco-Arroyave, J.R., Vargas-Bonilla, J.F., Nöth, E. (2016) Parkinson’s Disease Progression Assessment from Speech Using GMM-UBM. Proc. Interspeech 2016, 1933-1937, doi: 10.21437/Interspeech.2016-1122
@inproceedings{ariasvergara16_interspeech,
author={T. Arias-Vergara and J.C. Vasquez-Correa and Juan Rafael Orozco-Arroyave and J.F. Vargas-Bonilla and Elmar Nöth},
title={{Parkinson’s Disease Progression Assessment from Speech Using GMM-UBM}},
year=2016,
booktitle={Proc. Interspeech 2016},
pages={1933--1937},
doi={10.21437/Interspeech.2016-1122}
}