ZA African Next Voices
Data Science for Social Impact, University of PretoriaZA African Next Voices, also called Swivuriso, is a speech recognition dataset of about 3,000 hours of scripted and unscripted speech in seven South African languages: isiZulu, isiXhosa, Sesotho, Setswana, Xitsonga, isiNdebele and Tshivenda. Topics span agriculture, health, finance and general subjects, with speaker-disjoint splits. The South Africa Next Voices team publishes it under the dsfsi-anv Hugging Face organization, alongside a compressed copy.
Openness
3 high confidence- license
- cc-by-4.0(card adds a use restriction prohibiting TTS, voice cloning and voice synthesis)
- access
- auto(automatic Hugging Face gate)
- dataset_card
- present(languages, hours, fields, splits and use restriction described)
The audio carries an attribution-only license behind an automatic click-through, but the card forbids any use for speech synthesis or voice cloning, which narrows what that license alone would allow. A test portion is held back for future shared tasks.
- https://huggingface.co/api/datasets/dsfsi-anv/za-african-next-voices recorded 2026-09-24
"gated": "auto"; cardData license "cc-by-4.0".
- https://huggingface.co/datasets/dsfsi-anv/za-african-next-voices recorded 2026-09-24
"License Creative Commons Attribution 4.0 International (CC BY 4.0)"; "strictly prohibit any use of this dataset for any form of text-to-speech (TTS), voice cloning, voice synthesis"; test split "reserved for future shared tasks".
Adoption
2 high confidenceHugging Face downloads summed over the full repository and its Opus-compressed copy of the same audio. Downloads count file fetches behind an automatic gate, not users.
- https://huggingface.co/api/datasets/dsfsi-anv/za-african-next-voices recorded 2026-09-24
1102 downloads in the trailing 30 days for dsfsi-anv/za-african-next-voices
- https://huggingface.co/api/datasets/dsfsi-anv/za-african-next-voices-compressed recorded 2026-09-24
100 downloads in the trailing 30 days for dsfsi-anv/za-african-next-voices-compressed
Capability
3 medium confidenceZA African Next Voices gives seven South African languages scripted and unscripted speech transcribed by people, five hundred hours each for five of them, with a paper on its design, and its publishers' Whisper models and a community isiZulu recognizer are trained on it. Its roughly three thousand hours are more than any other open set here offers these languages, yet a small share of the hundreds of thousands behind the largest English speech corpora.
- https://arxiv.org/abs/2512.02201 recorded 2026-09-24
Abstract: "Swivuriso, a 3000-hour multilingual speech dataset developed as part of the African Next Voices project" with baseline ASR results.
- https://huggingface.co/datasets/dsfsi-anv/za-african-next-voices recorded 2026-09-24
Language table: isiZulu, isiXhosa, Sesotho, Setswana, Xitsonga 500 target hours, isiNdebele and Tshivenda 250, all "100%" released; models include dsfsi-anv/whisper-large-v3-turbo-anv-zul.
Verified 2026-09-24