Characteristics and Validity of Commercially Available Technologies Analyzing Voice Features to Assess Parkinson's Disease.

BACKGROUND

Interest in technologies for quantitative assessment of Parkinson's disease (PD) is growing, particularly those enabling voice and speech analysis. However, clinical validation varies widely, much algorithm data remain unpublished, and real-world use is still limited.

OBJECTIVE

The aim was to provide an overview of the characteristics and validity of commercially available tools that utilize machine learning to assess voice and speech features for identifying PD or assessing disease severity.

METHODS

The literature was reviewed to (1) identify, (2) compare, and (3) group commercially available tools deployed in smartphone apps and other scalable platforms based on their (1) previous use in PD, (2) feasibility, and (3) successful clinimetric testing. The International Parkinson and Movement Disorder Society Digital Tools and Technologies Repository; online databases, for example, PubMed; and web pages related to the entities associated with different technologies were searched for tools that use voice assessment. Identified devices were grouped into 3 categories: (1) "recommended," (2) "suggested," or (3) "listed."

RESULTS

Twelve relevant technologies were identified. One was placed in the "recommended" group ("Dysarthria Analyzer"), 6 in the "suggested" group, ("Audeering," "Aural Analytics," Canary Speech," "KI:Elements," "Modality.ai," and "Redenlab"), and 5 in the "listed" group ("i-Prognosis," "No Pa," and "PdAssist," as well as "mPower" and "HopkinsPD").

CONCLUSIONS

Despite their potential for clinical and research applications, their suitability in real-world environments remains to be determined.