Ama Branoa Banful is a graduating MPhil candidate in Computer Engineering at Kwame Nkrumah University of Science and Technology (KNUST) and a researcher with the Responsible Artificial Intelligence Lab (RAIL). She also serves as a Junior Field Officer on the Artificial Intelligence for Sustainable Development (AI4SD) project, where her work looks at how AI can create inclusive technologies for underrepresented communities.
She has successfully defended her MPhil thesis, which focuses on Automatic Speech Recognition (ASR) for low-resource languages, with particular attention to English-Twi code-switching. Code-switching, where speakers alternate between two or more languages during a conversation, is common in Ghana and other multilingual societies, and remains a significant issue in the current state of monolingual ASR systems, which are mostly trained on high-resource languages.
To address the lack of natural English-Twi datasets in this domain, Ama generated synthetic data using SpaCy-based text augmentation and blended English and Twi text-to-speech (TTS) models to create realistic code-switched text and audio. Experimentation with pre-trained ASR models, such as Whisper and Wav2Vec2, in zero-shot and few-shot scenarios, reveals the advantages and drawbacks of synthetic data in the progress of inclusive speech recognition in multilingual societies.
Her research offers valuable insights into how synthetic data can contribute to the development of more inclusive ASR systems in the area of code-switching, while also highlighting the importance of real-world code-switched datasets in future research.
In addition to her thesis, Ama has worked on projects, including the I-SEE project at RAIL, which develops AI-powered tools to assist individuals living with visual impairments in accessing support.
She credits the lab’s guidance, mentorship, and collaborative spirit for shaping her research journey. Her work reflects RAIL’s mission of advancing responsible, Africa-centred AI that addresses real needs in society and makes technology more inclusive.