Honest Signals in the Recognition of Functional Relational Roles in Meetings

In this paper we focus on the usage of honest signals for the automatic recognition of functional relational roles. We consider 16 speech honest signal features, along with three fidgeting visual features, and investigate their potential for role classification comparing: a) different labeling convention for relational roles, b) the contribution of honest signals’ class to classification, and c) independent vs. joint classification of task and social roles, in order to understand the extent to which the redundancy between the task and socio roles can be exploited for classification.