Articulation correctness measurement of Parkinson's disease using low resource-intensitve segmentation methods

Abstract

The paper is about speech fluency measurements and classification of patients with Parkinson's disease. Speech is a process that is controlled by a complex system in humans. Parkinson's disease also affects speech production. In this work, we examine the speech fluency of PD patients using two language-independent segmentation method: forward-backward divergence segmentation (FBDS) and transient-stationary segmentation (TSS). Significance tests show that most features are different among the two groups. Support vector machines were applied performing automatic classification tests. The highest achieved accuracy was 0.76 with 0.72 F1-score. This implies that the features calculated may help in differentiating PD from healthy speech and aid a decision support tool without the need of a complex, language-dependent ASR system.