Parkinson's Disease Detection Using Hybrid Siamese Neural Network and Support Vector Machine in Multilingual Voice Signal.

Parkinson's disease (PD) is a neurodegenerative disorder that affects speech in earlier stage as a motor functions due to dopamine depletion in the substantia nigra. Voice signals serve as key markers for early PD detection. Most of the studies of the voice change in PD are focused in particular language. However, limited studies address the issue in multilingual scenarios. Detection of PD in multilingual environments poses several challenges due to language-specific acoustic variations and unavailability of large dataset. To overcome these challenges, this study proposes a hybrid Siamese Neural Network (SNN) with support vector machine (SVM) for effective detection across multiple languages. The proposed study captures the patterns of PD and healthy in varied acoustic conditions of different languages by contrastive learning. The sustained vowel phonation /a/, /e/, /i/, /o/, and /u/ in Italian and the vowel /a/ in US-English and Spanish are used in the study. Experimental results demonstrate that the hybrid SNN-SVM achieved the highest accuracy of 93% in the vowel /o/ in the Italian dataset. In cross-language scenarios when model trained in Spanish and tested in US-English an accuracy of 82.7% is achieved. The hybrid approach enhances PD detection accuracy despite language variations, providing a potential solution for real-time scenario.