ALZHEIMER’S DETECTION FROM SPEECH USING SUPERVISED EMBEDDINGS WITH TEMPORAL ATTENTION
Health Informatics - An International Journal · 28.05.2026 · 15 (2) · 21-30
DOI: 10.5121/hiij.2026.15202 ↗Kurz gefasst
Fachartikel in der Zeitschrift Health Informatics - An International Journal, online erschienen am 28.05.2026. Verfasst von Jamshaid Iqbal und 5 weiteren Autor:innen.
Themen: Demenz
Abstract (Originalsprache)
Speech-based assessment of cognitive decline offers a low-burden and non-invasive alternative to resource-intensive diagnostic procedures. However, many existing systems rely on transcripts, multimodal fusion, or fixed acoustic features, which increase pipeline complexity and can limit practical deployment. This paper presents a transcript-free speech-only framework for three-way classification of Alzheimer’s disease (AD), mild cognitive impairment (MCI), and healthy controls (HC) from spontaneous speech. The proposed method combines supervised fine-tuning of a pretrained Wav2Vec2.0 encoder with lightweight auxiliary acoustic cues and temporal attention for recording-level prediction. Experiments on the NCMMSC-AD 2021 benchmark are conducted under a strict patient-disjoint hold-out protocol in both short-speech and long-speech settings. The proposed system achieves 85.19% accuracy and 85.69% macro-F1 in the short-speech setting, and 88.04% accuracy and 87.57% macro-F1 in the long-speech setting. These results show that task-adapted acoustic representation learning and temporal aggregation can provide strong transcript-free performance for cognitive-status classification.
Zitieren
APA
Iqbal, J., Rehmat, S., Li, J., Sabir, H., Malik, A., & Zafar, M. (2026). ALZHEIMER’S DETECTION FROM SPEECH USING SUPERVISED EMBEDDINGS WITH TEMPORAL ATTENTION. Health Informatics - An International Journal, 15(2), 21-30. https://doi.org/10.5121/hiij.2026.15202
Vancouver
Iqbal J, Rehmat S, Li J, Sabir H, Malik A, Zafar M. ALZHEIMER’S DETECTION FROM SPEECH USING SUPERVISED EMBEDDINGS WITH TEMPORAL ATTENTION. Health Informatics - An International Journal. 2026;15(2):21-30. doi:10.5121/hiij.2026.15202
BibTeX
@article{Iqbal2026,
author = {Iqbal, Jamshaid and Rehmat, Sadaf and Li, Jiyun and Sabir, Hira and Malik, Arslan and Zafar, Mehtab},
title = {ALZHEIMER’S DETECTION FROM SPEECH USING SUPERVISED EMBEDDINGS WITH TEMPORAL ATTENTION},
journal = {Health Informatics - An International Journal},
year = {2026},
volume = {15},
number = {2},
pages = {21--30},
doi = {10.5121/hiij.2026.15202}
}
EUVIMED