An Extended Dataset of Extracted Acoustic Featuresfor Voice-Based Detection of Parkinson’s Disease

Parkinson's disease (PD) is a progressive neurodegenerative disorder for which reliable, non-invasive biomarkers remain limited. Voice alterations, often present in the early stages of PD, are promising biomarkers for early detection. Existing public datasets remain a bottleneck to developing and validating early detection methodologies. We present an extended dataset integrating features from the Oxford Parkinson's Disease dataset and Iyer et al. This standardized dataset enables machine learning model training and evaluation, facilitates reproducibility, and supports exploration of novel voice-based biomarkers. We additionally provide analyses comparing feature distributions and model performance across individual and combined datasets, highlighting expected differences due to heterogeneous cohorts and feature extraction methods.

帕金森病（Parkinson's disease, PD）是一种进行性神经退行性疾病，目前可靠的非侵入性生物标志物（biomarker）仍较为匮乏。帕金森病早期常出现的语音改变，是用于早期检测的颇具前景的生物标志物。现有的公开数据集仍是开发与验证早期检测方法的瓶颈。本研究构建了一款扩展数据集，整合了牛津帕金森病数据集（Oxford Parkinson's Disease dataset）与Iyer等人研究中的特征。该标准化数据集可支持机器学习模型的训练与评估，提升研究可复现性，并助力基于语音的新型生物标志物的探索。此外，本研究还提供了针对单一数据集与合并数据集的特征分布、模型性能的对比分析，阐明了因异质性队列与特征提取方法差异所带来的预期差异。