Analyzing Vocal Tremor as an Acoustic Biomarker for Parkinson's Disease Using Sustained Vowels

This study investigates the contribution of vocal tremor-related acoustic features to the automatic detection of Parkinson's disease (PD) from sustained vowel phonations. While most voice-based approaches rely on conventional perturbation and spectral descriptors, tremor-related parameters remain underexplored despite their physiological relevance to neuromotor impairment. A comprehensive set of acoustic features was extracted, including jitter, shimmer, noise-related measures, nonlinear descriptors, and tremor features derived from low-frequency modulations of fundamental frequency and amplitude. Experiments were conducted on multiple datasets with distinct recording conditions, and classification was performed using a multilayer perceptron under subject-independent cross-validation. Results indicate that conventional features remain highly discriminative, but tremor-related descriptors consistently contribute complementary information and improve classification performance, achieving accuracy up to 98% when combined with traditional features. Although not sufficient alone, tremor features enhance both performance and physiological interpretability. These findings support the inclusion of vocal tremor descriptors in PD detection systems and highlight their potential as interpretable acoustic biomarkers across diverse recording conditions.