Parkinson’s disease (PD) is a neurodegenerative disorder marked by progressive motor and non-motor impairments, with hypokinetic dysarthria as a salient, measurable symptom. Although traditional diagnostics are accurate, their cost and invasiveness hinder widespread screening, especially in low-resource settings. This study investigates the cross-linguistic diagnosis of fine-tuned multilingual Audio Spectrogram Transformer (AST) models trained on Korean, English, Italian, and Slovakian speech. Each model was trained monolingually and evaluated in-language and cross-lingually. The Korean-trained model achieved the highest native F1-score (0.95) and best cross-lingual average F1-score (0.91), suggesting certain languages enable better generalization for broader diagnostic transfer.