REPORT Hovsepyan_Idiap-RR-02-2026/IDIAP Syllable-Level Features for Speech Pathology Detection: A Case Study of Parkinson?s Disease Hovsepyan, Sevada Magimai-Doss, Mathew neurocomputational models neurodegenerative speech disorders Parkinson's disease detection pathological speech detection syllable-level-features syllables EXTERNAL http://publications.idiap.ch/attachments/reports/2026/Hovsepyan_Idiap-RR-02-2026.pdf PUBLIC Idiap-RR-02-2026 2026 Idiap May 2026 In this paper we explore newly proposed syllable level feature extraction for speech pathology detection: focusing on the Parkinson's disease detection from speech. The method is inspired by the spectro-temporal representations used in the neuro-computational models of speech perception, and simply represents a standardized representation of the frequency content of the syllable-like segments. In this study, we demonstrate that this type of representation is powerful enough to successfully detect various speech pathologies from speech samples, independent of speech type, language, and dataset. Furthermore, our analyses show that these representations lead to explainable and interpretable results, and that changes in pathologies are related to both system- and source-related changes in speech production. Overall, syllable-level features suggest themselves as a simple but robust and explainable approach to understand physiological changes in speech production due to different pathologies, as well as to detect those pathologies from speech.