Breaking Data Efficiency Dilemma: A Federated and Augmented Learning Framework For Alzheimer's Disease Detection via Speech

Xiao Wei; Bin Wen; Yuqin Lin; Kai Li; Mingyang gu; Xiaobao Wang; Longbiao Wang; Jianwu Dang

cs.CLFeb 16, 2026

Breaking Data Efficiency Dilemma: A Federated and Augmented Learning Framework For Alzheimer's Disease Detection via Speech

Authors: Xiao Wei, Bin Wen, Yuqin Lin, Kai Li, Mingyang gu, Xiaobao Wang, Longbiao Wang, Jianwu Dang

Organizations: Tianjin Key Laboratory of Cognitive Computing and Application, College of Intelligence and Computing, Tianjin University, Tianjin, China · Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China · College of Computer and Data Science, Fuzhou University, Fuzhou, China · Huiyan Technology (Tianjin) Co., Ltd, Tianjin, China

Abstract

Early diagnosis of Alzheimer's Disease (AD) is crucial for delaying its progression. While AI-based speech detection is non-invasive and cost-effective, it faces a critical data efficiency dilemma due to medical data scarcity and privacy barriers. Therefore, we propose FAL-AD, a novel framework that synergistically integrates federated learning with data augmentation to systematically optimize data efficiency. Our approach delivers three key breakthroughs: First, absolute efficiency improvement through voice conversion-based augmentation, which generates diverse pathological speech samples via cross-category voice-content recombination. Second, collaborative efficiency breakthrough via an adaptive federated learning paradigm, maximizing cross-institutional benefits under privacy constraints. Finally, representational efficiency optimization by an attentive cross-modal fusion model, which achieves fine-grained word-level alignment and acoustic-textual interaction. Evaluated on ADReSSo, FAL-AD achieves a state-of-the-art multi-modal accuracy of 91.52%, outperforming all centralized baselines and demonstrating a practical solution to the data efficiency dilemma. Our source code is publicly available at https://github.com/smileix/fal-ad.

Figures & tables

Panel of Figure 1 — Figure 1: Overview of the proposed FAL-AD framework. (a) Cross-category Voice Conversion-based Data Augmentation: generates pathological speech samples by recombining speaker features from one category and content features from another. (b) Adaptive Federated Learning: clients collaboratively train under privacy constraints and select their optimal model from the federation history. (c) Cross-Modal Fusion Model: depicts the word-level alignment and attentive fusion process of acoustic and textual features.

Methods → Modals ↓		Previous Work (Centralized Paradigm)				Our Work
Methods → Modals ↓		C-Attn [ 22 ]	Ying [ 23 ]	Bang [ 24 ]	CogniAlign [ 15 ]	CL	CL+Aug	LL	LL+Aug	FL	FL+Aug
Audio	Acc	75.30	71.20	69.01	80.12	74.55	79.39	68.69	68.08	83.84	85.05
Audio	F1	76.00	73.10	70.39	79.46	73.39	79.14	65.68	67.10	83.67	84.64
Text	Acc	73.50	78.90	83.10	86.77	84.85	86.67	78.39	79.80	87.68	90.30
Text	F1	73.50	79.00	83.10	86.59	84.69	86.63	77.48	79.55	87.64	90.28
Both	Acc	77.20	83.70	87.32	90.36	86.06	86.67	78.59	80.61	89.70	91.52

Table 1: Performance comparison (Accuracy and F1-Score in %) across different modalities and learning paradigms on the ADReSSo dataset. Results from previous centralized methods are compared against our implementations under Centralized Learning (CL), Local Learning (LL), and Federated Learning (FL) paradigms, both with and without data augmentation (Aug). The best value in each group is marked with an underline, and the global best value across all groups is marked with bold. Note that CL here is the strict reproduction version of CogniAlign using its source code.

Algorithms	sFL		pFL		aFL
Algorithms	Acc	F1	Acc	F1	Acc	F1
FedAdam	9 0.91	9 0.88	90.00	89.97	90.91	90.88
FedAdagrad	89.70	89.58	88.79	88.74	90.00	89.95
FedYogi	87.27	87.26	88.48	88.40	88.79	88.71
FedProx	89.70	89.63	90.60	90.52	90.91	90.84
FedAvg	9 0.91	90.85	9 0.81	9 0.75	91.52	91.45

Table 2: Performance comparison (Accuracy and F1-Score in %) of federated learning algorithms with different model section strategy: standard (sFL), personalized (pFL), and adaptive (aFL) federated learning. The best performance across section strategies is bold, while the best performance across FL algorithms is underlined.

Speech-Based Dementia Detection

Speech-based dementia detection uses spoken language and vocal patterns to identify cognitive impairment and support accessible screening.

See all 44 papers

LSEAD: A Privacy-Preserving LLM-Based Speech Analysis Framework for Early Alzheimer's Disease Screening
Aug 7, 2026Xin Wang, Yingchao Huang, Yuhan Su +2Speech-Based Dementia Detection Alzheimer's Disease Classification
Faculty of Digital Innovation, Arts & Sciences, Saskatchewan Polytechnic, Regina SK S4S 5X1, Canada · School of Basic Medical Sciences, Hebei University, Baoding 071000, China · Department of Civil & Environmental Engineering and School of Mining & Petroleum Engineering, University of Alberta, Edmonton AB T6G 2H5, Canada +1
Cleaner Speech, Weaker Generalization: Revisiting Pitt-Derived Benchmarks for Alzheimer's Disease Detection
Aug 31, 2026Luqi Sun, Shreeram Suresh Chandra, Lin Zhang +5Speech-Based Dementia Detection Speech Enhancement
Center for Language and Speech Processing (CLSP), Johns Hopkins University · Research Center for Information Technology Innovation, Academia Sinica · University of Michigan, Ann Arbor +1
Robust Cross-Domain Speech-Based Alzheimer's Disease Detection via Iterative Adversarial Self-Training
Sep 12, 2026Luqi Sun, Shreeram Suresh Chandra, Aurosweta Mahapatra +3Speech-Based Dementia Detection Alzheimer's Disease Classification
Johns Hopkins University Baltimore, USA · University of Michigan Ann Arbor, USA · Carnegie Mellon University Pittsburgh, USA

Aug 7, 2026J/K move · Enter open · S save

LSEAD: A Privacy-Preserving LLM-Based Speech Analysis Framework for Early Alzheimer's Disease Screening

Xin Wang, Yingchao Huang, Yuhan Su, Shanshan Yao+1

Faculty of Digital Innovation, Arts & Sciences, Saskatchewan Polytechnic, Regina SK S4S 5X1, Canada · School of Basic Medical Sciences, Hebei University, Baoding 071000, China · Department of Civil & Environmental Engineering and School of Mining & Petroleum Engineering, University of Alberta, Edmonton AB T6G 2H5, Canada+1

Speech-Based Dementia Detection Alzheimer's Disease Classification

Open paper AskPDF arXiv

Early diagnosis of Alzheimer's disease (AD) is critical for enabling timely interventions that may slow disease progression and improve patient outcomes. There is a growing need for AD detection methods that are non-invasive and cost-effective, especially in real-world clinical settings with diverse patient populations and recording conditions. Speech-based screening addresses these needs by using natural speech collected without specialized equipment. Recent advances in large language models (LLMs) have improved speech analysis by providing rich linguistic representations and strong generalization. In this study, we propose LSEAD, a speech-based AD detection framework using pretrained open-source LLMs. Speech recordings are automatically transcribed, and text embeddings are extracted using locally deployed LLMs. Principal component analysis (PCA) is applied to reduce dimensionality before classification. Because the framework relies only on speech transcripts and locally deployed models, it supports privacy-preserving AD risk assessment without external data exchange. We evaluate LSEAD on the ADReSS20 and ADReSSo2021 benchmark datasets. Experimental results show that LLM-based embeddings generalize well across datasets and improve AD classification accuracy by up to 5 percent over existing methods, especially for early-stage detection. These results demonstrate that LSEAD provides a practical, secure, and scalable approach for early AD screening.

Abstract

Speech-Based Dementia Detection

Explore similar work

LSEAD: A Privacy-Preserving LLM-Based Speech Analysis Framework for Early Alzheimer's Disease Screening

Cleaner Speech, Weaker Generalization: Revisiting Pitt-Derived Benchmarks for Alzheimer's Disease Detection

Robust Cross-Domain Speech-Based Alzheimer's Disease Detection via Iterative Adversarial Self-Training