An Attention-Based Hybrid Network for Automatic Detection of Alzheimer’s Disease from Narrative Speech

Jun Chen, Ji Zhu, Jieping Ye


Alzheimer’s disease (AD) is one of the leading causes of death in the world and affects at least 50 million individuals. Currently, there is no cure for the disease. So a convenient and reliable early detection approach before irreversible brain damage and cognitive decline have occurred is of great importance. One prominent sign of AD is language dysfunction. Some aspects of language are affected at the same time or even before the memory problems emerge. Therefore, we propose an automatic speech analysis framework to identify AD subjects from short narrative speech transcript elicited with a picture description task. The proposed network is based on attention mechanism and is composed of a CNN and a GRU module. We obtained state-of-the-art cross-validation accuracy of 97 in distinguishing individuals with AD from elderly normal controls. The performance of our model makes it reasonable to conclude that our approach reveals a considerable part of the language deficits of AD patients and can help with the diagnosis of the disease from spontaneous speech.


 DOI: 10.21437/Interspeech.2019-2872

Cite as: Chen, J., Zhu, J., Ye, J. (2019) An Attention-Based Hybrid Network for Automatic Detection of Alzheimer’s Disease from Narrative Speech. Proc. Interspeech 2019, 4085-4089, DOI: 10.21437/Interspeech.2019-2872.


@inproceedings{Chen2019,
  author={Jun Chen and Ji Zhu and Jieping Ye},
  title={{An Attention-Based Hybrid Network for Automatic Detection of Alzheimer’s Disease from Narrative Speech}},
  year=2019,
  booktitle={Proc. Interspeech 2019},
  pages={4085--4089},
  doi={10.21437/Interspeech.2019-2872},
  url={http://dx.doi.org/10.21437/Interspeech.2019-2872}
}