Can an affect-sensitive system afford to be context independent?

Abstract

There has been a wave of interest in affect recognition among researchers in the field of affective computing. Most of these research use a context independent approach. Since humans may misunderstand other’s observed facial, vocal, or body behavior without any contextual knowledge, we question whether any of these human-centric affect-sensitive systems can be robust enough without any contextual knowledge. To answer this question, we conducted a study using previously studied audio files in three different settings; these include: no contextual indication, one level of contextual knowledge (either action or relationship/environment), and two levels of contextual knowledge (both action and relationship/environment). Our work confirms that indeed the contextual knowledge can improve recognition of human emotion.