Development of Emotion Rankers Based on Intended and Perceived Emotion Labels

Zhenghao Jin, Houwei Cao


In emotion datasets, intended emotion labels and perceived emotion labels both contain valuable information about how human express and perceive emotions, and there is a considerable mismatch between the two. In this paper, we propose a novel method to derive relative labels for preference learning using both the intended labels during emotion expression and the perceived labels given by all raters during perceptual evaluation. Based on analyzing the agreement between the intended and perceived labels, as well as the consistence among all perceptual ratings, we propose three pairwise ranking rules to generate multi-scale relevant scores for preference learning. We further build three sets of rankers for six basic emotions based on the three ranking rules. Through evaluation on the CREMA-D database, we demonstrate that, by considering both intended and perceived labels, our proposed rankers significantly outperform the rankers only relying on the perceptual ratings. We further combine the ranking scores of individual emotions for multi-class classification. Through experiments, we show that the emotion classification systems with ranking information significantly outperform the conventional SVM classifiers.


 DOI: 10.21437/Interspeech.2019-1831

Cite as: Jin, Z., Cao, H. (2019) Development of Emotion Rankers Based on Intended and Perceived Emotion Labels. Proc. Interspeech 2019, 3277-3281, DOI: 10.21437/Interspeech.2019-1831.


@inproceedings{Jin2019,
  author={Zhenghao Jin and Houwei Cao},
  title={{Development of Emotion Rankers Based on Intended and Perceived Emotion Labels}},
  year=2019,
  booktitle={Proc. Interspeech 2019},
  pages={3277--3281},
  doi={10.21437/Interspeech.2019-1831},
  url={http://dx.doi.org/10.21437/Interspeech.2019-1831}
}