Alzheimer disease speech detection improved across different settings
Robust Cross-Domain Speech-Based Alzheimer's Disease Detection via Iterative Adversarial Self-Training
Sound
Summary
Detecting Alzheimer's disease by analyzing speech is helpful but challenging when methods only work well on one specific dataset. The authors designed a new approach called Iterative Adversarial Self-Training (IAST) that helps the detection system work reliably even when speech data come from different places or conditions. They do this by teaching the system to focus on disease-related speech patterns and ignore unrelated differences between datasets. Their method improves accuracy when tested across various recording environments and speaker groups.
What this means in practice
- •For clinical data engineers: Improve speech-based Alzheimer’s detection tools to work reliably on data from multiple clinics with different recording settings.
- •For voice assistant developers: Enhance voice interfaces to better recognize health-related speech patterns despite varied user environments and microphones.
Authors
Luqi Sun, Shreeram Suresh Chandra, Aurosweta Mahapatra, Emily Mower Provost, Brian MacWhinney, Berrak Sisman
Abstract
As Alzheimer's disease (AD) has increasingly become a major global public health issue, speech-based AD detection has attracted widespread attention. However, most existing methods are trained and evaluated on a single dataset, often leading to severe cross-domain performance degradation due to reliance on dataset-specific artifacts rather than disease-related speech cues. In real-world applications, reliable Alzheimer's disease detection requires models that are robust to variations in recording environments, speakers and data collection conditions. To address this challenge, this paper adopts unsupervised domain adaptation to learn robust, domain-invariant feature representations in the absence of target-domain diagnosis labels. On this basis, a novel unsupervised domain adaptation method, Iterative Adversarial Self-Training (IAST), is proposed. Results demonstrate that IAST significantly improves the generalization ability and robustness under various cross-domain settings.