Comparing evaluation criteration for (automatic) emotion recognition

Owned to the fact that the automatic recognition of emotions still constitutes a hot topic in the research on adaptive human-computer interfaces, there exists a large and growing variety of approaches and systems presented in numerous publications each yielding a different performance at different capabilities. During the development of two approaches to speech-based emotion recognition and having tried to compare their performance with existing work, we have puzzled over the reliability and the explanatory power of the recognition rates given for different proposed systems. In this paper, we develop different approaches to assess the impact of an error occurring in the recognition of emotions leading to a larger variety of recognition rate definitions and we discuss their potential of comparability in application scenarios.