Alzheimer speech screening models struggle to work across settings
Why Alzheimer's Speech Screening Fails to Generalize: Bridging the Deployment Gap via Cross-Corpus Evidence Anchoring
Computation and LanguageSound
Summary
Detecting Alzheimer's using speech sounds helpful but current computer programs often fail when tried on different languages or recording setups. The authors studied four different sets of speech samples and found many speech features change in opposite ways depending on the dataset. They showed typical models drop in accuracy when tested across datasets and proposed a new method that mixes text analysis with carefully chosen speech clues to improve reliability across different situations. Their work points out the importance of checking how well features transfer and reporting results on the hardest cases.
What this means in practice
- •For mobile app developers: Build more reliable Alzheimer’s speech screening apps that work across languages and recording methods by using fusion of text and speech evidence anchors.
- •For clinical data teams: Audit and improve the reliability of speech-based cognitive assessment tools by evaluating feature consistency across datasets and reporting worst-case domain performance.
Authors
Zijian Lu, Sizhe Liu, Yin Zhang, Jixuan Deng, Xinrong Lin, Xinchen Yuan, Chicheng Jin, Yiping Zuo, Yuanchao Li
Abstract
Speech-based screening is a promising, non-invasive approach for detecting Alzheimer's disease and related cognitive risks. However, models trained on a single domain often generalize poorly to unseen languages, tasks, or recording protocols. This paper investigates this deployment gap using a leave-one-corpus-out evaluation across four distinct datasets. Among 70 interpretable speech and language features, 59 exhibit direction conflicts between healthy control and cognitive risk groups across corpora, with pause, silence, and speech rate showing high protocol sensitivity. Furthermore, while the XLM-R text baseline achieves strong average performance, its Area Under the ROC Curve (AUC) drops to 0.520 on the weakest held-out domain. A standard GroupDRO baseline reaches a 0.766 mean speaker AUC and a 0.504 worst-domain AUC under the same protocol. To address this, we propose a fusion method that integrates XLM-R text baseline scores with evidence anchors selected during training. Balanced fusion achieves a 0.785 mean speaker AUC, while anchor-heavy fusion raises the worst-case speaker AUC to 0.615. This work highlights the need to audit feature transferability and report worst-case domain robustness in cognitive speech screening.