Elamur Research⌕
☰
home / papers / 2026 / Evaluating Yoruba Text-to-Speech Systems for Accessible Computer-Based Testing in Visually Impaired Learners

Evaluating Yoruba Text-to-Speech Systems for Accessible Computer-Based Testing in Visually Impaired Learners

Paper · 2026 · Kausar Yetunde Moshood, Victor Tolulope Olufemi, Oreoluwa Boluwatife Babatunde et al.

Summary

Exams are increasingly delivered on computers, and this can exclude blind learners when assistive speech tools do not handle their language. This study asks whether current Yoruba text-to-speech (TTS) is good enough to read exam items to visually impaired students. The authors first rated four Yoruba TTS systems with Mean Opinion Scores. They kept the two most intelligible, Meta's mms-tts-yor and YarnGPT, and connected them to a custom web interface for computer-based testing (CBT). Five blind secondary students then answered the same ten WAEC Yoruba multiple-choice questions in three formats: read aloud by a fluent human speaker, in Braille, and through TTS.

Every participant ranked Braille first, citing control, quiet and independence. Human narration was clear but made students feel dependent on someone else. TTS came last because the speech sounded robotic, words were mispronounced and tones were handled poorly. The authors conclude that tonal, low-resource languages need tone-aware TTS built with input from users. They also call for better open Yoruba speech data to support fair digital assessment.

Key points

  • Pre-study MOS ratings (1–5) of four Yoruba TTS models:
    • MMS-TTS-Yor scored 3.7, the best result.
    • YarnGPT scored 3.2.
    • YorubaTTS scored 2.5 and was only partially intelligible.
    • Tacotron 2 scored 2.1.
  • MMS-TTS-Yor comes from Meta's Massively Multilingual Speech project and was trained on Yoruba Bible recordings. YarnGPT is a lightweight open-source model hosted on Hugging Face.
  • Participants were five visually impaired students (three male, two female) aged 15–18. All had Braille experience and little or no exposure to TTS.
  • The test material was ten four-option multiple-choice items from the 2024 WAEC Yoruba paper, stored in a CSV file. The items covered comprehension, tone-sensitive vocabulary and cultural expressions.
  • Preferences:
    • All five students chose Braille first.
    • Three found the human voice helpful.
    • Only one student liked the idea of TTS, and still criticised its quality and tonal errors.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Carnegie Mellon UniversityPittsburgh · United States539 works · 179,533 citations🎓Oregon State UniversityCorvallis · United States9 works · 907 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Evaluation & benchmarks · 1,220 worksSpeech & spoken language · 480 works

Cite

Cite as Kausar Yetunde Moshood, Victor Tolulope Olufemi, Oreoluwa Boluwatife Babatunde, Emmanuel Bolarinwa and Williams Oluwademilade. 2026. Evaluating Yoruba Text-to-Speech Systems for Accessible Computer-Based Testing in Visually Impaired Learners. Proceedings of the 7th Workshop on African Natural Language Processing (AfricaNLP 2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.africanlp-main.23

@inproceedings{moshood2026evaluating,
  title = {{Evaluating Yoruba Text-to-Speech Systems for Accessible Computer-Based Testing in Visually Impaired Learners}},
  author = {Kausar Yetunde Moshood, Victor Tolulope Olufemi, Oreoluwa Boluwatife Babatunde, Emmanuel Bolarinwa and Williams Oluwademilade},
  booktitle = {Proceedings of the 7th Workshop on African Natural Language Processing (AfricaNLP 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.africanlp-main.23},
  url = {https://doi.org/10.18653/v1/2026.africanlp-main.23}
}

References

20 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. American Foundation for the Blind. 2025. Screen readers. Accessed: 2025-02-22. find
  2. Rosana Ardila, Megan Branson, Kelly Davis, Michael Kohler, Josh Meyer, Michael Henretty, Reuben Morais, Lindsay Saunders, Francis Tyers, and Gregor Weber. 2020. Common Voice: A massively-multilingual speech corpus. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 421…
  3. Saheed Azeez. 2025. YarnGPT: Nigerian-accented English text-to-speech model. find
  4. Chika Chukwuma-Nosike and Peace Chukwuma. 2023. Computer-based test (CBT) evaluation innovation: Prospects in curriculum implementation in Nigeria. Journal of Agriculture and Food Sciences, 21(2). Special Issue: Man, Environmental Safety and Sustainability the Role of Research, Chapter 24, pp. 331–… find
  5. Manpreet Kaur Dhaliwal and Rohini Sharma. 2024. Improving accessibility and independence for blind/visually impaired persons based on speech synthesis technology. International Journal of Computer Applications, 186(28):12–20. find
  6. David Eberhard, Gary Simons, and Chuck Fennig. 2019. Ethnologue: Languages of the World, 22nd edition. SIL International. find
  7. Alexander Gutkin, Isin Demirsahin, Oddur Kjartansson, Clara E. Rivera, and Kólá Túbo.sún. 2020. Developing an open-source corpus of Yoruba speech. In Proc. of Interspeech 2020, pages 404–408, Shanghai, China. find
  8. ITU-T. 1996. Method for subjective determination of transmission quality. Technical Report Recommendation P.800, International Telecommunication Union - Telecommunication Standardization Sector. Retrieved from https://www.itu.int/rec/T-REC-P. 800-199608-I/en. ↗
  9. Jaehyeon Kim, Jungil Kong, and Juhee Son. 2021. Conditional variational autoencoder with adversarial learning for end-to-end text-to-speech. Preprint, arXiv:2106.06103. — [arXiv:2106.06103](https://arxiv.org/abs/2106.06103) find
  10. National Federation of the Blind. 2009. The Braille Literacy Crisis in America: Facing the Truth, Reversing the Trend, Empowering the Blind. NFB Jernigan Institute, Baltimore, MD. find
  11. Tolulope Ogunremi, Kola Tubosun, Anuoluwapo Aremu, Iroro Orife, and David Ifeoluwa Adelani. 2024. ‘Ir‘oy‘inSpeech: A multi-purpose Yorùbá speech corpus. Preprint, arXiv:2307.16071. — [arXiv:2307.16071](https://arxiv.org/abs/2307.16071) find
  12. Pawan Kumar Patel, Amey Karkare, and Gaurav Raheja. 2021. Inclusive accommodations for persons with visual impairments in computer-based tests. Studies in Health Technology and Informatics, 282:219–237. find
  13. Perkins School for the Blind. 2024. An overview of Braille devices. https://www.perkins.org/ resource/overview-braille-devices/. ↗
  14. Vineel Pratap, Adithya Tjandra, Bowen Shi, Jing Huang, Qiantong Xu, Aravindh Krishnaswamy Babu, and Michael Auli. 2023. Scaling speech technology to 1,000+ languages. Preprint, arXiv:2305.13516.
  15. Sirisha Rella. 2023. Exploring unique applications of text-to-speech technology. find
  16. Yi Ren, Yangjun Ruan, Xu Tan, Tao Qin, Sheng Zhao, Zhou Zhao, and Tie-Yan Liu. 2019. FastSpeech: Fast, robust and controllable text to speech. Preprint, arXiv:1905.09263. find
  17. Jonathan Shen, Ruoming Pang, Ron J. Weiss, Mike Schuster, Navdeep Jaitly, Zongheng Yang, Zhifeng Chen, Yu Zhang, Yuxuan Wang, R. J. Skerry-Ryan, Rif A. Saurous, Yannis Agiomyrgiannakis, and Yonghui Wu. 2018. Natural TTS synthesis by conditioning WaveNet on MEL spectrogram predictions. In 2018 IEEE I… find
  18. Kola Tubosun. 2023. Making machines speak Yorùbá. find
  19. Klá Túbsún and Adéday Olúòkun. 2017. Yorùbá text-to-speech system. https://www.ttsyoruba.com. Accessed: 2025-02-22. waec. The west african examinations council. Accessed: 2025-02-22. ↗
  20. Carmen Willings. 2017. Educational assessments for students who are blind or visually impaired. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/moshood-2026-evaluating-yoruba-text-speech-systems/