Swivuriso: ZA-African Next Voices

Swivuriso is unique in that it combines both scripted andunscripted speech, reflecting how people actually use language in daily life. All recordings are collected through ethical, community-centered processes, ensuring that participants are fairly engaged and that the data benefits the wider community. This approach strengthens both the quality of the dataset and its long-term impact.

The dataset covers the following seven languages, with the goal of building a balanced resource that reflects South Africa’s linguistic diversity:

• isiZulu – 500h
• isiXhosa – 500h
• Sesotho – 500h
• Xitsonga – 500h
• Setswana – 500h
• isiNdebele – 250h
• Tshivenda – 250h

In total, the dataset will reach 3,000 hours of high-quality, multilingual audio. These recordings will form the foundation for robust ASR models, helping to break literacy barriers, make digital content locally relevant, and accelerate innovation in South African language technologies.