Articulation and Empirical Mode Decomposition Features in Diadochokinetic Exercises for the Speech Assessment of Parkinson’s Disease Patients

Abstract

Speech impairments are one of the earliest manifestations in patients with Parkinson's disease. Particularly, articulation impairments related to the capability of the speaker to move the limbs and muscles of the vocal tract have been observed in the patients. Articulation deficits have been evaluated in the patients mainly using diadochokinetic exercises, which consist in the rapid repetition of syllables like /pa-ta-ka/. This study considered different features to model several aspects of the diadochokinetic exercises, including the capacity to start/stop the vocal fold vibration, the speech rate, and the regularity of the diadochokinetic task. Articulation features are combined with others that result from an empirical mode decomposition procedure, which have been recently used to model dysphonia in Parkinson's patients. The features are used to classify Parkinson's patients and healthy speakers, and to predict the dysarthria severity of the participants according to a clinical scale. According to the results, articulation features are able to classify the presence of the disease with an accuracy up to 76%, and to predict the dysarthria level of the speakers with a Spearman's correlation of up to 0.68.

Exemplar-Based Sparse Representations for Detection of Parkinson's Disease From Speech

“…The best accuracy (76%) obtained in this study for the vowel production task with the proposed-NSRC approach is the same as the one reported in [21]. Furthermore, the best accuracy (82.5%) obtained with the proposed-NSRC approach for the DDK task is better than the accuracy of 76% reported in [29]. Note that the authors of [21] have not evaluated their models using the DDK task, and the authors of [29] have not evaluated their models using the vowel production task.…”

“…Note that the authors of [21] have not evaluated their models using the DDK task, and the authors of [29] have not evaluated their models using the vowel production task. Therefore, in comparison with [21], [29], the proposed-NSRC approach provides comparable or better performance in discriminating healthy speakers from PD patients.…”

The Phonetic Footprint of Parkinson's Disease

“…With respect to study designs, most related works perform speech analysis on signals acquired from dedicated speech tasks, such as sustained vowels or diadochokinesis (DDK) tasks (rapid repetition of consonant-vowel clusters) [15,16,17,18]. The major advantage of this setup is the combination of exercise and data collection.…”