At the Kwame Nkrumah University of Science and Technology (KNUST), a research team led by Professor Jerry John Kponyo, and including Ama Branoa Banful, Juliet Arthur, and Kenneth Dotse, is developing new approaches to make speech recognition systems more inclusive and representative of real-world communication.

Their project focuses on a key challenge: building automatic speech recognition (ASR) systems capable of understanding code-switching and non-standard speech in low-resource environments.

Building models that reflect reality

To address the limitations of existing systems, the team is developing a unified ASR framework trained on how people actually speak.

Their approach is grounded in extensive data collection:

• over 100 hours of code-switched speech from 109 speakers 
• more than 10 hours of speech from individuals with impairments, with a target of 20 hours 

This dual focus is essential. By integrating both multilingual speech patterns and impaired speech, the project aims to ensure that AI systems are not only more accurate but also more inclusive.

The dataset is being made available through open platforms, contributing to broader research and innovation efforts.

Rethinking performance and evaluation

One of the key challenges identified by the team concerns how AI systems are evaluated.

Traditional metrics, such as word error rates, often fail to capture how well models perform in complex, real-world scenarios like code-switching. A system may appear accurate according to standard benchmarks, yet perform poorly in everyday use.

This highlights the need to develop new evaluation frameworks that better reflect linguistic diversity and practical usability.

Designing for accessibility and scale

A central objective of the project is to ensure that solutions are accessible in low-resource environments.

Rather than relying on high-performance infrastructure, the team is working to develop models that can run on edge devices such as mobile phones, making them usable by a wider population.

This approach reflects a broader commitment to designing AI systems that are not only technically advanced, but also practically deployable and socially relevant.

A vision for inclusive AI

Beyond its technical contributions, the project advances a broader vision of AI development.

The team emphasises three key priorities:

• efficiency, ensuring that AI tools are accessible in everyday contexts; 
• open innovation, promoting open-source approaches and local ecosystems; 
• inclusion by design, ensuring that people with different linguistic and physical capacities are fully represented. 

Rather than focusing on a divide between Global North and South, the project adopts the perspective of the Global Majority, those who are currently underserved by existing AI systems.

Looking ahead

The team will present its findings at the AI for Good Global Summit in Geneva (7-10 July 2026) as part of ITU’s Kaleidoscope sessions, on Thursday, 9 July 2026, from 15:30 to 16:30. More information soon.

Their work highlights a crucial insight:
AI systems must be designed not only for performance, but for the diversity of human experience.

About AI for the Global Majority

AI for the Global Majority (AI4GM) is a joint initiative of the Geneva Graduate Institute, Microsoft, and the International Telecommunication Union (ITU) dedicated to supporting innovative, evidence-based, and context-sensitive research on how artificial intelligence can benefit the world’s majority populations.

Bringing together interdisciplinary teams from across regions and sectors, the initiative explores practical pathways for more inclusive, responsible, and impactful AI in areas such as governance, education, health, finance, and digital innovation.

Selected teams will present their work in Geneva as part of the AI for Good Global Summit, contributing to international discussions on the future of AI and global development.

By grounding innovation in local realities and lived experience, this project offers a powerful model for how AI can become more inclusive, accessible, and globally relevant.