Speech Signal Recognition for Parkinson's Disease Diagnosis Based on Heterogeneous Broad Balanced Domain Adaptation

The small sample size and high aliasing charact-eristics of Parkinson's disease (PD) speech data poses significant challenges in the application of machine learning methods in PD diagnosis. Despite the progress made by various transfer learning methods in enhancing the accuracy of PD speech recognition, the majority of these approaches rely on the assumption of a same feature space between the source and target domains. This study proposes a novel domain adaptation algorithm named Hetero-geneous Broad Balanced Domain Adaptation (HBBDA) to address the inconsistency of feature spaces between the source and target domains. Firstly, align the source domain data with the target domain data using broad learning theory. Secondly, it considers both the knowledge of the source and target domain data, and utilizes a small amount of labeled target domain data to guide the training of the source domain data. Experimental results show that HBBDA achieves an accuracy of 85.23% and a sensitivity of 92.49% on the target domain, indicating that it not only has high accuracy but also exhibits high sensitivity to patients, thus better assisting medical professionals in decision-making.