Abstract

In this paper we discuss both the speaker dependent and the listener dependent aspects in the assessment of emotions in speech. These dependencies form a basis to improve current emotion recognition systems as they can be applied in man-machine interaction, for instance. Emotion recognition in speech has gained much attention in recent years [1, 2, 3]. However, human evaluation of emotions has not been exploited to its full extent so far. Each person perceives emotions differently. Therefore in this paper we analyze evaluation variations and propose to use them to get better references for the emotion recognition systems. Also we investigate different speakers’ emotion expression variabilities which could be used to build speaker models for model-based speakerindependent automatic emotion recognition. For the analysis of variations in producing and perceiving emotions, we use a 3D emotion space …

Date

January 1, 1970

Authors

Michael Grimm, Kristian Kroschel, Shrikanth Narayanan

Journal

FORTSCHRITTE DER AKUSTIK

Volume

32

Issue

1

Pages

335