Swivuriso: The South African Next Voices Multilingual Speech Dataset

The Swivuriso dataset, comprising 3000 hours of multilingual speech, has been released to enhance automatic speech recognition (ASR) technologies for seven South African languages. It includes diverse topics such as agriculture and healthcare, and the paper outlines the dataset's design principles, ethical considerations, and baseline results from training ASR models, demonstrating its superiority over existing datasets for these languages. This resource is crucial for practitioners aiming to improve ASR performance in underrepresented languages and domains.