African Next Voices Kenya

license

cc-by-4.0(stated in the paper, Section 7

access

manual(users must share contact information and accept conditions on Hugging Face)

dataset_card

no("No dataset card yet" on all five repositories

The paper releases the audio under an attribution-only license, but each language repository asks for contact details and acceptance of conditions before download. The repositories themselves say nothing about license or composition.

• https://arxiv.org/html/2604.08448 recorded 2026-09-24

  "The AfriVoices-KE dataset is publicly available on Hugging Face under a CC BY 4.0 license, with access managed through a request form to track usage."

• https://huggingface.co/api/datasets/Anv-ke/Dholuo recorded 2026-09-24

  "gated": "manual"; no cardData license.

• https://huggingface.co/datasets/Anv-ke/Dholuo recorded 2026-09-24

  "You need to agree to share your contact information to access this dataset"; "No dataset card yet".

Hugging Face downloads summed over the five per-language repositories, each behind a gate that collects contact details. Downloads count file fetches, not approved users.

• https://huggingface.co/api/datasets/Anv-ke/Dholuo recorded 2026-09-24

  844 downloads in the trailing 30 days for Anv-ke/Dholuo

• https://huggingface.co/api/datasets/Anv-ke/Kalenjin recorded 2026-09-24

  267 downloads in the trailing 30 days for Anv-ke/Kalenjin

• https://huggingface.co/api/datasets/Anv-ke/kikuyu recorded 2026-09-24

  599 downloads in the trailing 30 days for Anv-ke/kikuyu

• https://huggingface.co/api/datasets/Anv-ke/Maasai recorded 2026-09-24

  248 downloads in the trailing 30 days for Anv-ke/Maasai

• https://huggingface.co/api/datasets/Anv-ke/Somali recorded 2026-09-24

  198 downloads in the trailing 30 days for Anv-ke/Somali

AfriVoices-KE gives five Kenyan languages scripted and spontaneous speech from thousands of native speakers with a detailed paper, and recognizers such as an East African model and a Kalenjin Whisper model are trained on it, though its repositories lack a full dataset card. Its roughly three thousand hours match its South African sibling, ZA African Next Voices, and are a small share of the hundreds of thousands in the largest English speech corpora.

• https://arxiv.org/abs/2604.08448 recorded 2026-09-24

  Abstract: "approximately 3,000 hours of audio across five Kenyan languages... 750 hours of scripted speech and 2,250 hours of spontaneous speech, collected from 4,777 native speakers".

• https://huggingface.co/datasets/Anv-ke/Kalenjin recorded 2026-09-24

  "Models trained or fine-tuned on Anv-ke/Kalenjin" lists leophill/w2v-bert-2.0-afrivoices-ea-6l-asr and Rekody/whisper-large-v3-turbo-kalenjin.