Tshivenda speech dataset

What is Africa Next Voices?

Africa Next Voices (ANV) is a large-scale initiative supported by the Gates Foundation and a network of research and technology partners to expand high-quality speech datasets for African languages. In South Africa, the project was coordinated by the Data Science for Social Impact (DSFSI) group at the University of Pretoria. Way With Words acted as the data production and workflow partner — designing and running recording, transcription, proofing, and quality control to deliver the South African component.

Read more about our journey building African Next Voices

How was this data collected?

The South African ANV dataset combines scripted and unscripted speech. Contributors were recruited from across the country and trained to record in their first language. Recordings were transcribed, proofed, and quality-checked by language specialists. The result is thousands of hours of ethically collected, community-driven speech that reflects how people actually speak — not scraped or synthetic sources.

How can I use this dataset?

The full multi-language dataset (Swivuriso) is available on Hugging Face. You can load data by language (e.g. isiZulu, isiXhosa, seSotho). Use restrictions apply: the data is not licensed for text-to-speech, voice cloning, or voice synthesis. For research, ASR, and language model training, see the dataset card and license on Hugging Face for full terms.

Who contributed to this project?

Thousands of South Africans — recorders, proofreaders, and language assistants — gave their time and voices to build this resource. We honoured participants with personalised certificates and fair compensation. For a contributor’s perspective on what it meant to be part of ANV, read Beyond the Data; for how we recognised everyone involved, see Honouring the Individuals Who Made Africa Next Voices Possible.