Overview

• The Hefei Institutes of Physical Science team led by Prof. LI Hai introduced CTCAIT, an end-to-end framework that analyzes speech to detect dysarthria linked to neurological disease.
• The system pairs a large-scale audio feature extractor with InceptionTime and cross-time and cross-channel attention to capture temporal and multivariate patterns.
• In evaluations, it achieved 92.06% accuracy on a local Mandarin dataset and 87.73% on an external English dataset, indicating cross-linguistic generalizability.
• The authors report at least a 2.17 percentage-point gain over earlier Mandarin results and include interpretability analyses and task comparisons showing structured speech can be more effective.
• The findings are published in Neurocomputing, and the team frames the work as guidance for future clinical screening and monitoring requiring wider real-world validation.