Swahili speech emotion recognition datasets

KuralHub lists 1 speech emotion recognition dataset for Swahili. Each entry below gives the emotion labels, size, recording setup, licence, where to get it and how to cite it. Swahili is also part of the benchmark. The best of 11 models reached 83.0% test accuracy (wavlm-base-plus).

Language family: Niger-Congo (Bantu)
ISO code: sw
Spoken mainly in: Tanzania, Kenya, Uganda, Rwanda, Burundi, the Democratic Republic of Congo, and parts of Mozambique, Malawi, and Somalia
Datasets listed: 1
In benchmark: Yes

Datasets at a glance

Dataset	Type	Samples	Speakers	Recording	Access	Licence
Kenyan Swahili SER	Audio			Studio / Acted / Spontaneous	On request

Benchmark results

Swahili 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 61.5% (median 69.0%). The best run reached 83.0% with wavlm-base-plus. Compare with other languages.

**Kenyan Swahili SER** (run ID `Swahili-Dataset`). Scores on the test split, best run highlighted.
Model	Family	Val acc.	Test acc.	Macro-F1	Weighted-F1	Logs
hubert-base-ls960	HuBERT	0.75	0.81	0.78	0.81	logs
hubert-large-ls960-ft	HuBERT	0.61	0.60	0.59	0.59	logs
wav2vec2-base	wav2vec 2.0	0.65	0.72	0.69	0.71	logs
wav2vec2-large-960h	wav2vec 2.0	0.37	0.44	0.43	0.44	logs
wav2vec2-large-lv60	wav2vec 2.0	0.41	0.35	0.26	0.24	logs
wav2vec2-xls-r-300m	wav2vec 2.0	0.35	0.33	0.33	0.33	logs
wav2vec2-xls-r-1b	wav2vec 2.0	0.35	0.45	0.41	0.42	logs
wavlm-base-plusbest	WavLM	0.80	0.83	0.80	0.83	logs
wavlm-large	WavLM	0.78	0.74	0.74	0.74	logs
whisper-small	Whisper	0.71	0.69	0.69	0.69	logs
whisper-large	Whisper	0.78	0.80	0.79	0.79	logs

Kenyan Swahili SER

Type: Audio
Emotion labels: Sad, Happy, Surprised, Angry, Calm
Size: Samples: 300
Recording: Studio / Acted / Spontaneous
Access: On request
Licence: Not stated
File format: .wav
Website: Dataset Homepage
Contact: Gregory Mikuro

TBD

Used in the KuralHub benchmark (run ID Swahili-Dataset). See the results above.

Citation

@misc{mikuro2023swahiliSER,
  author       = {Gregory Mikuro},
  title        = {Swahili Speech Emotion Recognition System},
  year         = {2023},
  howpublished = {\url{https://github.com/estherdatascientist/Swahili-Speech-Emotion-Recognition-System}},
  note         = {GitHub repository},
  url          = {https://github.com/estherdatascientist/Swahili-Speech-Emotion-Recognition-System}
}

Source: datasets/Swahili/Kenyan_Swahili.md · Suggest an edit

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

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