A decade of advances in speech and voice-based biomarkers for Alzheimer’s disease and dementia: a scoping review

Alzheimer’s disease and related dementias are major causes of disability and dependency in aging populations. Recent advances in artificial intelligence and digital signal processing suggest that speech and voice characteristics may serve as potential biomarkers of cognitive decline. This scoping review examines machine-learning approaches applied to speech and voice analysis for screening, monitoring, and staging Alzheimer’s disease and related dementias. Following PRISMA–ScR guidelines, database searches were conducted in PubMed, Scopus, IEEE Xplore, and ACM Digital Library in September 2025. Studies were included if they applied AI, machine learning, or signal processing to speech data from individuals with Alzheimer’s disease, mild cognitive impairment, or related dementias. A total of 39 studies met inclusion criteria. Feature extraction methods ranged from handcrafted acoustic features to self-supervised embeddings (wav2vec 2.0, HuBERT), and models included support vector machines, random forests, convolutional neural networks, and transformers. Reported classification performance varied substantially across datasets, speech tasks, feature extraction approaches, and validation protocols, with several studies reporting accuracies between 70% and 95% under controlled experimental settings. The review also found that the use of deep learning and self-supervised speech representation models increased after 2020. Studies using these methods often reported higher performance, although differences in datasets, tasks, and evaluation methods make direct comparisons difficult. However, substantial challenges remain, including limited external validation, dataset heterogeneity, inconsistent reporting practices, and reduced model interpretability. This scoping review maps emerging methodological trends and highlights key gaps that must be addressed to support the development of reproducible, multimodal, and clinically interpretable speech and voice biomarkers for cognitive impairment screening and monitoring.