A novel digital endpoint for detecting symptom changes in vocal tremor and speech prosody associated with Parkinson's Disease

The aim of this work is to develop measurement concepts to objectively quantify the severity of Parkinson’s Disease (PD) related speech symptoms, and to identify vocal abnormalities specifically associated with the prodromal stage. Although multiple approaches to this problem have been proposed in recent literature [1 - 8] in addition to commercially available speech analytics platforms (Winterlight Labs and Aural Analytics), there is currently no established product which incorporates the disparate aspects of affected speech to fully characterize Parkinson’s symptom progression, particularly in the prodromal phase. Our approach will utilize a custom smartphone-based speech assessment tool to extract multiple hypothesis-driven acoustic features from patient speech in a real life environment. The resultant features will be used to train a pair of supervised machine learning models; a regressor to predict clinical PD symptom severity scores, and a classifier to distinguish prodromal PD patients from both healthy matched controls and PD patients in more advanced phases of disease progression.