Automatic Assessment of Alzheimer's across Three Languages Using Speech and Language Features

Paula Andrea Pérez-Toro, Tomás Arias-Vergara, Franziska Braun, Florian Hönig, Carlos Andrés Tobón-Quintero, David Aguillón, Francisco Lopera, Liliana Hincapié-Henao, Maria Schuster, Korbinian Riedhammer, Andreas Maier, Elmar Nöth, Juan Rafael Orozco-Arroyave

Abstract: With the increasing prevalence of Alzheimer's Disease (AD) worldwide, it is essential to develop non-invasive methods to monitor the progression of the disease. Speech and language analyses are suitable for detecting the cognitive impairment of AD patients; thus, by analyzing changes in speech patterns and language use, researchers can develop methods to monitor AD remotely. In this paper, we investigated several speech and language techniques commonly used for the automatic detection of AD. Furthermore, we considered speech recordings of 448 patients in three different languages: Spanish (57), German (205), and English (186). Cross-lingual analysis was carried out using two classification approaches: (1) training/testing in one or more languages and (2) training in one language and testing in another. We obtained unweighted average recall values of up to 83% to classify AD using the first classification approach and up to 70% with the second.