Towards a voice-based severity scale for Parkinson’s disease monitoring

Abstract The unified Parkinson’s disease rating scale, used to monitor the disease progression, is based on visual assessments of motor symptoms. Vocal manifestations of Parkinson’s disease differ from the motor ones, specifically in their rate of change with disease severity. As such, a different scale is needed to provide the voice measures of the disease severity. This study employed a dataset of voice-quality features from repeated recordings of Parkinson’s disease patients. The changes of all voice features across the categories were evaluated using one-way analysis-of-variance and support vector regression. Significant changes and marked non-linearly increasing or decreasing trends were shown for all features, for the three-categories scale. Significant changes and trends were obtained in the 12-categories scale, but only for the mild category and the severe category range of scores. The findings imply a potential for voice-based monitoring for the early and late severity stages of Parkinson’s disease that could be continuously used by patients and provide timely warnings of deterioration.