Towards developing robust multimodal databases for emotion analysis

Understanding emotions can make the difference between succeeding and failing during communication. Several systems have been developed in the field of Affective Computing in order to understand emotions. Recently these systems focus into multimodal emotion recognition. The basis of each of these systems is emotion databases. Even though a lot of attention has been placed in capturing spontaneous emotion expressions, building an emotion database is a task with several challenges that are commonly neglected, namely: quality of the recordings, ground truth, multiple device recording, data labeling and context. In this paper we present a new spontaneous emotion database, with human-computer and human to human interactions. This database is composed by eight different synchronized signals, in four interaction tasks. Strategies on how to deal with emotion database construction challenges are explained in detail.