UNILAG Student Develops AI-Powered Text-to-Speech Model with Nigerian Accent
Saheed Azeez, a final-year Mechanical Engineering student at UNILAG, has developed YarnGPT, an AI-powered text-to-speech model that supports Nigerian accents.
Chris Oyeoku Okafor 
A final-year student of Mechanical Engineering at the University of Lagos (UNILAG), Saheed Azeez, has successfully developed an innovative text-to-speech AI model named YarnGPT. Unlike conventional models that primarily support European or American accents, YarnGPT is uniquely designed to accommodate Nigerian accents and languages, making it a pioneering effort in localized artificial intelligence applications.
Azeez's journey into technology began during his first year at UNILAG when he realized that programming, particularly Python, was essential for excelling in Mechanical Engineering. His interest in AI and machine learning intensified during the COVID-19 lockdown in 2020, leading him to embark on various projects.
Inspired by the lack of text-to-speech models with African language support and a personal setback during a tech job interview, Azeez decided to build YarnGPT. He collected data from Nigerian movies, extracting audio and transcriptions to create a model that produces speech in clear, Nigerian-accented English.
Speaking on the challenges faced during development, Azeez highlighted issues with data quality, transcription inconsistencies, and the cost of acquiring Graphics Processing Units (GPUs) necessary for training the model. Despite these obstacles, he successfully built YarnGPT, spending around $150 of his savings to fund the project.
The model has potential applications in audiobooks and education, enabling students to learn through localized speech synthesis. Although Azeez has received inquiries about commercializing YarnGPT, he currently intends to keep it open-source, allowing it to benefit as many people as possible.
Reflecting on his experience, Azeez expressed pride in his achievement and affirmed his desire to continue working in AI rather than traditional mechanical engineering. His project has gained significant attention, and he is already working on an improved version, set for release in the coming weeks.
The Vice-Chancellor of UNILAG, Prof. Folasade Ogunsola, OON, FAS, congratulated Saheed Azeez on his groundbreaking innovation, commending his dedication and wishing him success in his future endeavors.
