AfriBED: a Speech Recognition Bias Evaluation Dataset for Low Resource African Languages
Paper • 2610.26821 • Published
Datasets:
You need to agree to share your Hugging Face information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
I agree to use this data with the license conditions and use restrictions.
Log in or Sign Up to review the conditions and access this dataset content.
AfriBED: African speech Bias Evaluation Dataset
AfriBED is a speech recognition evaluation dataset to examine speaker attribute bias in automatic speech recognition (ASR) models for low-resource African languages.
The current release contains read speech data for Bemba and Nyanja.
Speaker Attributes: Gender and Speaker type
Complete Sets
| Language | Utterances | Speakers | Duration |
|---|---|---|---|
| Bemba | 3,928 | 80 | 9.32 hours |
| Nyanja | 6,260 | 127 | 18.53 hours |
| Total | 10,188 | 207 | 27.85 hours |
Mini Sets
| Language | Utterances | Speakers | Duration |
|---|---|---|---|
| Bemba Mini | 1,396 | 28 | 3.13 hours |
| Nyanja Mini | 1,198 | 24 | 3.57 hours |
| Total | 2,594 | 52 | 6.70 hours |
Citation
If you use AfriBED in your research, please cite as follows:
@inproceedings{sikasote-etal-2026-afribed,
title = {Afri{BED}: a Speech Recognition Bias Evaluation Dataset for Low Resource African Languages},
author = {Sikasote, Claytone and Suleman, Hussein and Buys, Jan},
booktitle = {2026 IEEE Spoken Language Technology Workshop Demo Track},
year = {2026},
publisher = {2026 IEEE Spoken Language Technology Workshop},
url = {https://openreview.net/forum?id=cPvecAsOA7},
}
- Downloads last month
- -