On January 23, 2025, Azeez Saheed, a University of Lagos undergraduate and rising AI innovator, unveiled YarnGPT via an X post. This text-to-speech (TTS) model, designed to generate Nigerian-accented English, aims to bridge the gap between global AI technologies and local cultural relevance.
He was previously known for his earlier project, Naijaweb, a Nigerian-centric dataset. Together, these projects are transforming the AI landscape in Nigeria, creating opportunities for more localized and contextually accurate AI solutions.
Read also: ChatGPT users frustrated due to ‘bad gateway’ error
What are Naijaweb and YarnGPT?
Azeez first made waves with Naijaweb, a dataset comprising 230 million GPT-2 tokens sourced from Nairaland, one of Nigeria’s largest online forums. This rich repository includes conversations on politics, entertainment, and everyday life, providing an authentic representation of Nigerian culture and speech patterns. By training AI models on this dataset, Saheed sought to improve their understanding of Nigerian Pidgin, local slang, and cultural nuances—elements often overlooked in global datasets.
Building on the success of Naijaweb, Azeez developed YarnGPT, a TTS model capable of generating natural-sounding Nigerian-accented English. With a range of voices like Idera, Chinenye, and Osagie, the model provides culturally relevant audio for applications in education, customer service, and entertainment. Unlike many TTS models that rely on complex external adapters, YarnGPT uses pure language modelling to deliver high-quality speech synthesis, making it both efficient and impactful.
Why does localised AI matter?
Global AI models like ChatGPT are often trained on datasets that fail to capture non-Western populations’ linguistic and cultural intricacies. This creates a gap in how these technologies serve diverse communities, including Nigeria.
At the core of YarnGPT is the drive to make AI more relatable and culturally relevant for Nigerians. Unlike generic text-to-speech models, YarnGPT captures the nuances of Nigerian-accented English, delivering authentic and familiar speech. This has significant implications for accessibility, enabling more Nigerians to engage with AI-driven content in education, media, and customer service.
By reflecting the diversity of Nigerian voices and speech patterns, YarnGPT enhances the cultural connection between users and technology. It also empowers businesses to communicate more effectively with local audiences, making AI interactions feel personal and meaningful.
Read also: Adebayo Ogunlesi joins OpenAI’s board of directors
A vision for Nigeria’s AI ecosystem
Azeez’s work is a testament to the untapped potential within Nigeria’s tech community. By leveraging local platforms like Nairaland and combining them with cutting-edge AI technologies, he is building tools that are not only innovative but also deeply relevant to the Nigerian experience.
With plans to scale up model sizes, support local languages, and even enable voice cloning, Azeez sets the stage for a future where AI technologies are accessible, inclusive, and truly representative of Nigeria’s diverse culture.
As AI continues to evolve, innovators like Azeez Saheed remind us of the importance of local talent and localized data in shaping a more equitable technological future.