aiqbench analysis
Swivuriso introduces a 3,000-hour multilingual speech dataset designed to benchmark automatic speech recognition (ASR) systems across seven South African languages, specifically addressing domain gaps in agriculture and healthcare. By detailing its ethical collection procedures and baseline model performance, the work provides a critical resource for improving ASR equity in underrepresented linguistic contexts. This release offers a practical benchmark for researchers aiming to expand low-resource language capabilities beyond standard English-centric datasets.
Swivuriso: The South African Next Voices Multilingual Speech Dataset
arXiv:2512.02201v3 Announce Type: replace Abstract: 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.
Abstract: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.
Work in Progress. Updated in June 2026
Computation and Language (cs.CL)
arXiv:2512.02201 [cs.CL]
(or arXiv:2512.02201v3 [cs.CL] for this version)
Full text under the publisher's open licence · 106 words