Detecting Disruptive Vocalizations for Ambient Assisted Interventions for Dementia

People suffering from dementia exhibit abnormal behaviors that can put them at risk or burden their relatives and caregivers. Many of these behaviors have acoustic manifestations, such as shouting, mumbling, cursing or making repetitive tapping. In this paper we propose an approach for detecting disruptive behavior manifested through audio. The solution proposed is an specialized speech, verbal and ambient sound detector and classifier, using speech and CASA techniques. We illustrate how the detection of these symptoms can be used to enact non-pharmacological interventions aimed at stopping such behavior or mitigating its negative impact.