Publication Details
Enhancing multilingual recognition of emotion in speech by language identification
Matějka Pavel, Ing., Ph.D. (DCGM)
GAVRYUOKOVA, M.
Povolný Filip, Ing.
MARCHI, E.
SCHULLER, B.
multilingual emotion recognition, language identification,language families
We investigate, for the first time, if applying model selectionbased on automatic language identification (LID) can improvemultilingual recognition of emotion in speech. Six emotionalspeech corpora from three language families (Germanic, Romance,Sino-Tibetan) are evaluated. The emotions are representedby the quadrants in the arousal/valence plane, i. e., positive/negative arousal/valence. Four selection approaches forchoosing an optimal training set depending on the current languageare compared: within the same language family, acrosslanguage family, use of all available corpora, and selectionbased on the automatic LID. We found that, on average, theproposed LID approach for selecting training corpora is superiorto using all the available corpora when the spoken languageis not known.
@inproceedings{BUT163404,
author="SAGHA, H. and MATĚJKA, P. and GAVRYUOKOVA, M. and POVOLNÝ, F. and MARCHI, E. and SCHULLER, B.",
title="Enhancing multilingual recognition of emotion in speech by language identification",
booktitle="17TH ANNUAL CONFERENCE OF THE INTERNATIONAL SPEECH COMMUNICATION ASSOCIATION - Proceedings (INTERSPEECH 2016)",
year="2016",
journal="Proceedings of Interspeech",
number="9",
pages="2949--2953",
publisher="International Speech Communication Association",
address="San Francisco",
doi="10.21437/Interspeech.2016-333",
issn="1990-9772",
url="https://www.isca-speech.org/archive/Interspeech_2016/pdfs/0333.PDF"
}