Detecting emotional traits in call centre interactions can be beneficial to the quality management of the services provided, since this reveals the positioning of both speakers, i.e. satisfaction or frustration and anger on the customers' side, and stress detection, disappointment mitigation or failure to provide the requested service on the operators' side. This paper describes a machine learning approach to classify emotional speech units occurring in a call centre dataset by employing emotion-related labels, automatically extracted acoustic features as well as additional context-related features.