Ensemble methods for spoken emotion recognition in call-centres
The recognition of human emotions is a very important task towards implementing more natural computer interfaces. A good annotation of the emotional corpora employed by researchers is fundamental to optimize the performance of the emotion recognizers developed. In this paper we discuss several aspects to be considered in order to obtain as much information as possible from this kind of corpora, and propose a novel method to include them automatically during the annotation procedure. The experimental results show that considering information about the usersystem interaction context, as well as the neutral speaking style of users, yields a more fine-grained human annotation and can improve machine-learned annotation accuracy by 24.52%, in comparison with the classical annotation based on acoustic features.