KuralHub lists 2 speech emotion recognition datasets for Hungarian. Each entry below gives the emotion labels, size, recording setup, licence, where to get it and how to cite it. Hungarian is also part of the benchmark. The best of 11 models reached 52.0% test accuracy (hubert-base-ls960).
Language family
Uralic (Ugric)
ISO code
hu
Spoken mainly in
Hungary, Serbia (in Vojvodina), Austria (in Burgenland), Slovenia (in Prekmurje), Romania (in Transylvania),
Hungarian 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 36.7% (median 43.0%). The best run reached 52.0% with hubert-base-ls960. Compare with other languages.
HESD (run ID HungarianEmotionalSpeechCorpus). Scores on the test split, best run highlighted.
This dataset comprises recordings of three Hungarian sentences, each articulated to express eight fundamental emotions: happiness, sadness, anger, fear, surprise, disgust, contempt, and neutral. These recordings were performed by nonprofessional speakers. The primary objective of this dataset is to facilitate the examination of human emotion perception and to support the development of automatic emotion recognition systems. To achieve this, the dataset was utilized in perception tests comparing human recognition of emotions and in training Hidden Markov Models (HMM) for automatic emotion recognition, optimizing parameters such as acoustic preprocessing and the number of states in the models.
Used in the KuralHub benchmark (run ID HungarianEmotionalSpeechCorpus). See the results above.
Citation
@inproceedings{toth2008speech,
title={Speech emotion perception by human and machine},
author={T{\'o}th, Szabolcs Levente and Sztah{\'o}, David and Vicsi, Kl{\'a}ra},
booktitle={Verbal and Nonverbal Features of Human-Human and Human-Machine Interaction: COST Action 2102 International Conference, Patras, Greece, October 29-31, 2007. Revised Papers},
pages={213--224},
year={2008},
organization={Springer}
}
The SEWA dataset is a multimodal dataset designed for emotion recognition and analysis. It includes audio and video data collected from participants during natural interactions. The dataset aims to study emotional states in real-world scenarios, focusing on cross-cultural differences and spontaneous emotional expressions. Data was collected using wearable devices and audio-visual recordings, making it a rich resource for affective computing research.
Citation
@article{kossaifi2019sewa,
title={Sewa db: A rich database for audio-visual emotion and sentiment research in the wild},
author={Kossaifi, Jean and Walecki, Robert and Panagakis, Yannis and Shen, Jie and Schmitt, Maximilian and Ringeval, Fabien and Han, Jing and Pandit, Vedhas and Toisoul, Antoine and Schuller, Bj{\"o}rn and others},
journal={IEEE transactions on pattern analysis and machine intelligence},
volume={43},
number={3},
pages={1022--1040},
year={2019},
publisher={IEEE}
}
Know another Hungarian speech emotion dataset, or a newer version of one listed here? Suggest it or open a pull request.
This page is generated from datasets/Hungarian/ in the KuralHub repository.
If you use a dataset, cite its authors. If KuralHub helped you find it, please also cite the KuralHub paper.