High-Accuracy Voice-Based Classification Between Patients With Parkinson’s Disease and Other Neurological Diseases May Be an Easy Task With Inappropriate Experimental Design

TLDR

Recent voice cepstral analysis studies report up to 90 % accuracy in distinguishing Parkinson’s disease from other neurological disorders. Our experiments show that while voice cepstral analysis can achieve high accuracy on healthy speakers, it is unreliable for distinguishing neurological diseases due to confounding by speaker traits such as gender and age, underscoring the need for stricter reporting assumptions.

Abstract

Recently, based on voice cepstral analysis, (Benba et al, 2016) have reported discrimination between patients with Parkinson's disease and different neurological disorders with high classification accuracy up to 90%. Using disorders with high classification accuracy up to 90%. Using the same approach, we were able to experimentally separate two groups of normal healthy speakers with 96% classification accuracy and showed that the method proposed by Benba et al. may not be appropriate for discrimination between different neurological diseases. In particular, voice cepstral analysis appears to be sensitive to specific speakers' characteristics such as gender or age. Our findings emphasize several assumptions that can be considered as basic necessary conditions for research reporting speech data in progressive neurodegenerative diseases.