Incorporating Contextual Audio for an Actively Anxious Smart Home

In this paper we explore an emotive, multi-modal smart house. The smart house is an instance of a monitoring application, inspired by the need to provide semi-autonomous assisted living for elderly and infirm people. A particular aspect of smart environments relevant to the care of the elderly is the detection of potential hazards. A hazardous situation represents an abnormal activity or event. Consequently, to detect abnormality we model normality, that is, the normal activities associated with a user's interaction with the environment. We use the concept of anxiety as a measure of normality modelled with a probabilistic approach. The anxiety is associated with a hazardous device using a fusion of multi-model data. The data is gathered from simple sensors, and from information derived from the audio domain indicating the presence of an activity within the environment. We present the results for the anxiety for a number of activity sequences, both normal and abnormal. The pervasive nature of the audio data enabled the detection of activity when interactions between a user and device didn't occur, successfully preventing false hazardous situations from being detected.