Affective behaviour could provide an indicator of Alzheimer’s disease and help develop clinical tools for automatically detecting andmonitoring disease progression. In this paper, we present a study of the predictive value of emotional behaviour features automaticallyextracted from spontaneous speech using an affect recognition system for Alzheimer’s dementia detection. The effectiveness of affectivebehaviour features for Alzheimer’s Disease detection was assessed on a gender and age balanced subset of thePitt Corpus, a spontaneousspeech database from DementiaBank. The affect recognition system was trained using the extended Geneva Minimalistic AcousticParameter Set (eGeMAPS) and the Berlin database of emotional speech. The output of this system provides classification scores or classposterior probabilities of 6+1 emotions as an input for statistical analysis and Alzheimer’s dementia detection. The statistical analysisshows that the non-AD subjects have higher mean value of classification scores forangeranddisgust, along with a higher entropyof classification scores than AD subjects. The AD subjects have a higher classification scores for thesademotional behaviour thannon-AD. This paper also introduces a novel ‘affective behaviour representation’ feature vector for Alzheimer’s dementia recognition.Results show that classification models based solely on affective behaviour attain 63.42% detection accuracy.
|Title of host publication||LREC: Resources and ProcessIng of linguistic, para-linguistic and extra-linguistic Data from people with various forms of cognitive/psychiatric/developmental impairments (RaPID)|
|Editors||Dimitrios Kokkinakis, Kristina Lundholm Fors, Charalambos Themistocleous, Malin Antonsson, Marie Eckerström|
|Publisher||European Language Resources Association (ELRA)|
|Publication status||Published - 11 May 2020|