DFKI-LT - Multidimensional meaning annotation of listener vocalizations for synthesis
Multidimensional meaning annotation of listener vocalizations for synthesis
3 Proceedings of the Workshop Emotion and Computing 2011, Berlin, Germany, Springer, 2011
Listener vocalizations convey affective and epistemic states behind the listeners intentions while the interlocutor is talking. The meaning annotation of such vocalizations is a crucial step in synthesis of listener vocalizations. This paper presents a perception study to annotate meaning of vocalizations. In this study, subjects annotate (characterize) a set of listener vocalizations using a multi-dimensional set of meaning descriptors. The set of stimulus vocalizations is selected based on intonation clustering. We investigate the typical impressions and the appropriateness of meanings conveyed by vocalizations, based on high agreement ratings provided by the participants. We also discuss the suitability of the annotation procedure to generate expressive listener vocalizations.
Files: BibTeX, Pammi_etal_2011.pdf