University of Lagos Student Launches YarnGPT to Bring Authentic Nigerian Voices to AI Text-to-Speech

From a Student Project to National Recognition

Saheed Azeez, a final-year Mechanical Engineering student at the University of Lagos, has transitioned his open-source text-to-speech model into a fully-fledged public application. Originally developed as an open-source project, YarnGPT quickly gained significant public attention, capturing the interest of Nigeria's Minister of Communications, Innovation and Digital Economy, Dr. 'Bosun Tijani. The success of the initial model helped Azeez secure a high-paying professional role, but his ultimate goal remained creating a practical system capable of solving communication challenges for everyday users.

The Technical Architecture of YarnGPT

Unlike conventional text-to-speech architectures that rely on external adapters, YarnGPT is built on pure language modeling. Released under the Apache-2.0 license, the model utilizes a Llama-based architecture with transformers and safetensors. Developers can access the open-source repository on GitHub under the username saheedniyi02, or deploy the model locally via Hugging Face. The system leverages tools such as the WavTokenizer and the outetts library to synthesize high-quality, culturally relevant speech without complex structural additions.

A Complete Suite for Local Languages

The newly launched web platform, available at yarngpt.ai, is designed to accommodate the unique pronunciations, nuances, and complexities of Nigerian languages. The system supports Yoruba, Igbo, Hausa, Pidgin, and Nigerian-accented English. Currently free to use, the web application allows users to input text directly, paste webpage URLs, or upload documents up to 100 megabytes in size. Users can select from various voice characters, such as Idera, to generate natural-sounding audio.

Expanding Use Cases for Content Creators and Developers

Beyond basic text-to-speech synthesis, the YarnGPT platform offers several advanced features for content creators, educators, and businesses. Users can translate and dub videos from English into native Nigerian languages, generate audiobooks, and create voiceovers for social media. For web publishers, the platform provides an embeddable audio reader for blogs, while software developers can access a dedicated API to integrate YarnGPT directly into their own applications.

What this means for Africa: The launch of YarnGPT shows how local developers can build tailored, open-source AI models that preserve indigenous languages and make digital content accessible to millions of native speakers.