Afrikaans speech emotion recognition datasets

KuralHub lists 1 speech emotion recognition dataset for Afrikaans. Each entry below gives the emotion labels, size, recording setup, licence, where to get it and how to cite it. Afrikaans is also part of the benchmark. The best of 11 models reached 71.0% test accuracy (whisper-small).

Language family: Indo-European (Germanic)
ISO code: af
Spoken mainly in: South Africa, Namibia
Datasets listed: 1
In benchmark: Yes

Datasets at a glance

Dataset	Type	Samples	Speakers	Recording	Access	Licence
AfrikaansSE	Audio	800		Studio (Local Youtube Channel)	On request	Creative Commons Attribution license

Benchmark results

Afrikaans 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 52.4% (median 59.0%). The best run reached 71.0% with whisper-small. Compare with other languages.

**AfrikaansSE** (run ID `AfrikaansEmotionalSpeechCorpus`). Scores on the test split, best run highlighted.
Model	Family	Val acc.	Test acc.	Macro-F1	Weighted-F1	Logs
hubert-base-ls960	HuBERT	0.65	0.67	0.62	0.66	logs
hubert-large-ls960-ft	HuBERT	0.32	0.41	0.41	0.41	logs
wav2vec2-base	wav2vec 2.0	0.57	0.66	0.67	0.67	logs
wav2vec2-large-960h	wav2vec 2.0	0.38	0.43	0.41	0.41	logs
wav2vec2-large-lv60	wav2vec 2.0	0.36	0.40	0.40	0.40	logs
wav2vec2-xls-r-300m	wav2vec 2.0	0.38	0.37	0.37	0.37	logs
wav2vec2-xls-r-1b	wav2vec 2.0	0.35	0.24	0.15	0.11	logs
wavlm-base-plus	WavLM	0.53	0.59	0.60	0.60	logs
wavlm-large	WavLM	0.29	0.63	0.58	0.61	logs
whisper-smallbest	Whisper	0.75	0.71	0.70	0.70	logs
whisper-large	Whisper	0.68	0.65	0.63	0.63	logs

AfrikaansSE

Creation of an Afrikaans speech corpora for Speech Emotion Recognition

Type: Audio
Emotion labels: Anticipation, Joy, Trust, Fear, Surprise, Sadness, Disgust and Anger
Size: Number of samples: 800
Recording: Studio (Local Youtube Channel)
Access: On request
Licence: Creative Commons Attribution license
File format: .wav
Paper: Paper Link
Contact: mail1, mail2

This dataset was created to support research in automatic speech emotion recognition (SER) for the Afrikaans language. It comprises high-quality audio recordings captured in a controlled studio environment. The recordings include both acted and spontaneous speech, covering eight emotional categories: Anticipation, Joy, Trust, Fear, Surprise, Sadness, Disgust, and Anger. The corpus is designed to facilitate the development and evaluation of SER systems tailored to Afrikaans, providing researchers with a valuable resource that reflects natural as well as expressive emotional speech patterns.

Used in the KuralHub benchmark (run ID AfrikaansEmotionalSpeechCorpus). See the results above.

Citation

@inproceedings{norval2022creation,
  title={Creation of an afrikaans speech corpora for speech emotion recognition},
  author={Norval, Michael and Wang, Zenghui},
  booktitle={2022 2nd International Conference on Robotics, Automation and Artificial Intelligence (RAAI)},
  pages={193--196},
  year={2022},
  organization={IEEE}
}

Source: datasets/Afrikaans/AfrikaansSE.md · Suggest an edit

Know another Afrikaans speech emotion dataset, or a newer version of one listed here? Suggest it or open a pull request.

This page is generated from datasets/Afrikaans/ in the KuralHub repository. If you use a dataset, cite its authors. If KuralHub helped you find it, please also cite the KuralHub paper.