KuralHub lists 1 speech emotion recognition dataset for Korean. Each entry below gives the emotion labels, size, recording setup, licence, where to get it and how to cite it. Korean is also part of the benchmark. The best of 11 models reached 81.0% test accuracy (whisper-small).
Korean is one of the 29 languages in the KuralHub benchmark. Each of the 11 frozen speech models was trained with a small classification head on the shared emotions and scored on a held-out test split. Over all 11 runs the mean test accuracy was 57.7% (median 70.0%). The best run reached 81.0% with whisper-small. Compare with other languages.
KESDy18 (run ID KESDy18). Scores on the test split, best run highlighted.
The Korean Emotional Speech Dataset (KESDy18) was collected in 2018 for speech emotion recognition research. It contains recordings of 30 voice actors uttering 20 sentences in four emotions (“angry,” “happy,” “neutral,” “sad”). Six annotators labeled each segment with one of seven categorical emotions, arousal (1–5), and valence (1–5). Labels were determined by majority vote and averaged scores. The dataset includes recordings from two microphones: a cellphone’s built-in mic (PM) and an external mic (EM). The published version contains KESDy18_EM (Shure S35) data.
Used in the KuralHub benchmark (run ID KESDy18). See the results above.
Citation
@article{noh2021multi,
title={Multi-path and group-loss-based network for speech emotion recognition in multi-domain datasets},
author={Noh, Kyoung Ju and Jeong, Chi Yoon and Lim, Jiyoun and Chung, Seungeun and Kim, Gague and Lim, Jeong Mook and Jeong, Hyuntae},
journal={Sensors},
volume={21},
number={5},
pages={1579},
year={2021},
publisher={MDPI}
}
Know another Korean speech emotion dataset, or a newer version of one listed here? Suggest it or open a pull request.
This page is generated from datasets/Korean/ in the KuralHub repository.
If you use a dataset, cite its authors. If KuralHub helped you find it, please also cite the KuralHub paper.