unza-speech-lab/zambezi-voice

Repository for multilingual speech data resources for native languages of Zambia. ## Zambezi Voice ### 1. Introduction The Zambezi Voice project is an on-going effort by the University of Zambia speech and language research group to develop/create speech and language data resources that would enable and foster research and development of language technology systems for under-resourced native languages of Zambia. ### 2. Objective To build speech and language data resources for under-resourced languages of Zambia that will: - enable the development of speech and language technologies: - Speech Recognition (ASR) - Machine Translation (MT) - Speech Translation (ST) - Multilingual Speech Recognition - serve as benchmark for academic and industry tools on Zambian languages. The long-term goal is to curate language data resources resources for all seventy-two (72) languages spoken in Zambia. In the medium term, we are focussing on the seven (7) main local languages spoken in Zambia: Bemba, Nyanja, Tonga, Lozi, Kaonde, Lunda and Luvale. ### 3. Datasets #### 3.1. Labelled Datasets [Read speech styled] Item Lang Code Files Hours Speakers Male Female Tasks 1 Nyanja nya 9167 25 12 3 7 ASR 3 Tonga toi 9354 22 9 5 4 ASR 2 Lozi loz 2924 6 6 4 2 ASR 4 Bemba bem 15500 26 18 10 8 ASR 5 Kikaonde kqn - - - - - ASR 6 Lunda lun - - - - - ASR 7 Luvale lue - - - - - ASR *Last Updated: 8/10/2024 #### 3.2. Unlabelled Audio Collections [Radio broadcast styled] Item Lang Code Audio Files Hours Audio Segments Hours Link 1 Nyanja nya 26 25 6976 10 Download 2 Tonga toi 122 101 38012 60 Download 3 Lozi loz 37 30 8845 15 Download 4 Bemba bem 533 162 26855 63 Download 5 Lunda lun 50 39 13424 20 Download *Last Updated: 31/01/2023 ### 4. Team - Claytone Sikasote [Lead] - Mofya Phiri - Mayumbo Nyirenda, PhD ### 5. Citation If you use this speech dataset in your project or research, please consider citing as follows: @inproceedings{sikasote23_interspeech, author={Claytone Sikasote and Kalinda Siaminwe and Stanly Mwa …