A Hybrid Deep Learning Framework for Alzheimer’s Detection Through Voice Biomarkers

Alzheimer's disease (AD) is the most common form of dementia, leading to progressive cognitive, linguistic, and behavioral decline.Traditional diagnostic methods such as cognitive tests, neuroimaging, and clinical evaluations are often expensive, invasive, and time consuming.Recent advancements in artificial intelligence have enabled the use of speech as a non-invasive and cost effective biomarker for AD detection.Prior studies have explored both linguistic and acoustic features, and more recently, hybrid models combining these modalities.However, many of these approaches rely on complex transcription pipelines or language-dependent data, limiting their scalability and generalization.Building upon this foundation, we propose an enhanced hybrid audio based framework that integrates handcrafted acoustic features with deep YAMNet embeddings, effectively capturing both lowlevel signal characteristics and high level semantic representations.The fused feature set is used to train multiple machine learning classifiers and a deep neural network (DNN) for automatic classification of AD and non-AD participants.Experimental evaluations on publicly available datasets achieved an 89% testing accuracy, surpassing prior hybrid and single-feature approaches.These findings highlight the effectiveness of the proposed model and demonstrate the potential of speech as a scalable, language independent biomarker for Alzheimer's detection.