In recent years, a growing number of corpora have been created for the analysis of the speech uttered by patients with Parkinson’s disease, yet only a fraction are openly accessible to the research community. This paper presents a systematic review and comparative analysis of all known publicly available speech corpora developed for this purpose, regardless of language or collection context. The study identifies ten datasets, which are individually evaluated across five key dimensions: (1) demographics and size; (2) recording protocols and quality; (3) speech task types; (4) accessibility and licensing; and (5) availability of raw audio and/or acoustic features. The evaluation reveals considerable variability in recording conditions, task diversity, metadata richness, and openness. Some datasets prioritise clinical precision and controlled environments, whereas others emphasise ecological validity and scalability by relying on mobile or telephonic recordings. The trade-offs involved are highlighted, and critical gaps in linguistic diversity, longitudinal follow-up, and multimodal integration are identified. By mapping the current landscape of accessible PD speech datasets, this work supports reproducibility, benchmarking, and informed dataset selection for future studies. It also provides concrete recommendations for the development of new corpora aligned with open science practices and clinical applicability.