Korean Emotional Speech Dataset (KESDy18)
The Korean Emotional Speech Dataset
(KEDy18) is collected for the research of speech emotion recognition in 2018.
The KESDy18 comprises speech samples in which 30 voice actors uttered 20
sentences while expressing the four given emotions of “angry”, “happy”,
“neutral”, and “sad”. The six external taggers evaluated the speech segments
while listening to the recorded utterances. The annotators tagged one of seven
categorical emotion labels (“angry”, “sad”, “happy”, “disgust”, “fear”,
“surprise”, “neutral”). And, they tagged labels of arousal (low-high: 1-5) and
valence-level (negative-positive: 1-5) on a 5-point scale for each segment. The
final categorical emotion label was determined by a majority vote. The label of
arousal and the valence-level were determined from the average value of the
levels tagged by the evaluators. KESDy18 simultaneously collected speech data
from two heterogeneous microphones (i.e., a cell phone's built-in microphone
(PM) and an external microphone (EM) connected to a computer). The published
dataset is the speech data of KESDy18_EM (Shure S35).
Funding
o
This work was supported by Electronics and
Telecommunications Research Institute (ETRI) grant funded by the Korean
government. [18ZS1100, Core Technology Research for Self-Improving Artificial
Intelligence System].
Files
o Speech files spoken by 30 voice actors (.wav / 2880
files)
o
Utterance sentence files (.txt / 2880 files)
o
Emotion label file (.xlsx / 1 file)
How to download the KESDy18
o
Log in to the site, agree to the contents of the End User License Agreement for permission to use this dataset.
o
If the data manager allows the download through this
site, it can be downloaded.
Publications
All documents
and papers that report on research that use any of the KESDy18 will acknowledge
this by citing the following papers:
Noh, K.J.; Jeong, C.Y.;
Lim, J.; Chung, S.; Kim, G.; Lim, J.M.; Jeong, H. Multi-Path and
Group-Loss-Based Network for Speech Emotion Recognition in Multi-Domain
Datasets. Sensors 2021, 21, 1579. https://doi.org/10.3390/s21051579
