Abstract: Early detection of dementia is essential for enabling timely clinical intervention, slowing disease progression, and improving overall patient care and quality of life. Conventional diagnostic methods such as neuropsychological assessments, brain imaging techniques, and clinical evaluations are often expensive, time-consuming, and dependent on specialized expertise, making them less accessible in rural or resource-limited settings. In response to these challenges, this paper presents an AI-based tool for early-stage dementia detection using speech analysis as a non-invasive and cost-effective alternative. Speech is a natural and information-rich medium that reflects cognitive processes, and subtle impairments in memory, attention, and executive functioning often manifest in acoustic, prosodic, and linguistic patterns during spontaneous speech production. The proposed system extracts a comprehensive set of features from recorded speech samples, including acoustic features such as Mel-Frequency Cepstral Coefficients (MFCCs), pitch, jitter, shimmer, and spectral properties; prosodic features such as speech rate, pause frequency, and pause duration; and linguistic features such as lexical diversity, syntactic complexity, word frequency distribution, and semantic coherence. The methodology follows a structured pipeline consisting of speech preprocessing (noise reduction, normalization, silence removal), feature extraction, feature selection to remove redundant and irrelevant attributes, model training, and performance evaluation. Multiple machine learning and deep learning algorithms are implemented and compared, including Support Vector Machines (SVM), Random Forest classifiers, and Long Short-Term Memory (LSTM) networks, which are particularly effective in modeling sequential and temporal dependencies in speech data. The models are trained and validated using appropriate cross-validation techniques to ensure robustness and generalization. Performance metrics such as accuracy, sensitivity, specificity, precision, recall, and F1-score are used to evaluate classification effectiveness in distinguishing early dementia cases from healthy controls. Experimental results demonstrate that the proposed AI-based tool achieves high diagnostic performance, highlighting the effectiveness of integrating acoustic and linguistic features for cognitive assessment. The findings suggest that speech-based AI systems can function as reliable, scalable, and remote screening solutions, supporting clinicians in early diagnosis and enabling proactive intervention strategies for individuals at risk of cognitive decline.

Keywords: Early Dementia Detection, Speech Analysis, Artificial Intelligence, Machine Learning, Acoustic Features

Downloads: | DOI: 10.17148/IARJSET.2026.13365