Explainable Artificial Intelligence and Multilingual Voice Analysis for Early Parkinsons Disease Detection: A Comprehensive Review and Proposed Framework

Parkinson’s disease (PD) is a progressive neurodegenerative disorder with everincreasing worldwide prevalence imposing important clinical and socioeconomic burdens. Where motor symptoms, the target of conventional diagnosis, occur relatively late in disease, speech and voice changes have become an earlier feature. These patterns of reduced pitch variation, monotonous prosody, jitter, shimmer, breathiness and imprecise articulation indicate laryngeal control dysfunction. Machine learning (ML) and deep learning (DL) enable automatic PD detection using acoustic features and neural architectures. However, challenges persist: lack of interpretability, linguistic bias from English-dominant datasets, and vulnerability to recording variability. This paper reviews advances in voice-based PD detection, emphasizing Explainable AI (XAI) and multilingual modeling. We also present a framework integrating classical ML, deep models, self-supervised encoders, and interpretability tools such as SHAP, LIME, integrated gradients, and attention maps.