Efficiency of Voice Features Based on Consonant for Detection of Parkinson's Disease

Rekha P M Viswanathan • Parham Khojasteh • Behzad Aliahmad • Sridhar Arjunan • Sanjay Ragnav • Peter Kempster • Kitty Wong • Jennifer Nagao • Dinesh Kant Kumar

10:00 - 17:00 | Monday 29 October 2018 | Foyer



The objective of the study is to determine the efficiency of features extracted from sustained voiced consonant /m/ in the diagnosis of Parkinson’s Disease (PD). The diagnostics applicability of the phonation /m/ is also compared with that of sustained phonation /a/, the one which is commonly employed in PD speech studies. The study included 40 subjects out of which 18 were PD and 22 were controls. The features extracted were used in SVM classifier model to differentiate PD and healthy subjects. The phonation /m/ yielded classification accuracy of 93% and Matthews Correlation Coefficient (MCC) of 0.85 while the classification accuracy for phonation /a/ was 70% and MCC of 0.39. The spearman correlation coefficient analysis also showed that the features from /m/ phonation were highly correlated with the Unified Parkinson’s Disease Rating Scale (UPDRS-III) motor score. The results suggest the applicability of features corresponding to nasal consonant in the diagnosis and progression monitoring of PD.