This paper introduces Swivuriso, a 3000-hour multilingual speech dataset developed as part of the African Next Voices project, to support the development and benchmarking of automatic speech recognition (ASR) technologies in seven South African languages. Covering agriculture, healthcare, and general domain topics, Swivuriso addresses significant gaps in existing ASR datasets. We describe the design principles, ethical considerations, and data collection procedures that guided the dataset creation. We present baseline results of training/finetuning ASR models with this data and compare to other ASR datasets for the langauges concerned.
@article{arxiv.2512.02201,
title = {Swivuriso: The South African Next Voices Multilingual Speech Dataset},
author = {Vukosi Marivate and Kayode Olaleye and Sitwala Mundia and Andinda Bakainga and Unarine Netshifhefhe and Mahmooda Milanzie and Tsholofelo Hope Mogale and Thapelo Sindane and Zainab Abdulrasaq and Kesego Mokgosi and Chijioke Okorie and Nia Zion Van Wyk and Graham Morrissey and Dale Dunbar and Francois Smit and Tsosheletso Chidi and Rooweither Mabuya and Andiswa Bukula and Respect Mlambo and Tebogo Macucwa and Idris Abdulmumin and and Seani Rananga},
journal= {arXiv preprint arXiv:2512.02201},
year = {2026}
}