Voice analysis for early dementia detection using neural networks

In this study, we present an approach to identify voice markers for early dementia detection. Our method involves extracting vocal features from speech and analyzing them using class-dependent principal component analysis to eliminate redundant variables and create meaningful variables. The resulting variables are fed into a machine learning algorithm to obtain a probabilistic prediction of the presence of dementia.We developed a noninvasive, low cost, and side-effects free approach. The obtained results showed the model has a high potential to detect dementia with accuracy=0.972, precision=0.983, recall=0.968, and F1-score=0.975.