Emotion recognition from speech using multi-classifier systems and RBF-ensembles

Abstract

This work provides a detailed overview of related work on the emotion recognition task. Common definitions for emotions are given and known issues such as cultural dependencies are explained. Furthermore, labeling issues are exempli-fied, and comparable recognition experiments and data collections are introduced in order to give an overview of the state of the art. A comparison of possible data acquisition methods, such as recording acted emotional material, induced emotional data recorded in Wizard-of-Oz scenarios, as well as real-life emotions, is provided. A complete automatic emotion recognizer scenario comprising a possible way of collecting emotional data, a human perception experiment for data quality benchmarking, the extraction of commonly used features, and recognition experiments using multi-classifier systems and RBF ensembles, is included. Results close to human performance were achieved using RBF ensembles, that are simple to implement and trainable in a fast manner. © 2008 Springer-Verlag Berlin Heidelberg.