Abstract

Parkinson’s disease (PD) is a progressive neurodegenerative condition characterized by dopamine depletion in the substantia nigra, which manifests through motor impairments and distinct vocal alterations. Recognizing temporal speech variations as critical biomarkers, this study proposes an optimized Convolutional Neural Network (CNN) framework for the early and precise detection of PD. Utilizing a dataset comprising 195 individuals (147 PD patients and 48 healthy controls), the research implements an advanced preprocessing pipeline involving signal denoising, feature normalization, and Bayesian hyperparameter optimization to refine classification efficiency. Experimental results demonstrate that the CNN classifier achieves a superior diagnostic accuracy of 96.85%, complemented by high sensitivity and a robust F1 score. The model maintains consistent predictive reliability even under suboptimal acoustic conditions, highlighting its capability to identify disease-specific vocal signatures effectively. These findings underscore the potential of the proposed deep learning architecture as a scalable clinical tool for early PD screening, facilitating timely therapeutic interventions and improved patient management within integrated diagnostic workflows.