In order to assess the quality of the corpus, we resorted to human observers for evaluating both the eliciting movie clips (Section V-A), and the acquired data in the form of videos containing renderings of the 3-D template tracking coupled with the original audio signal (Section V-B).