South African graduate develops AI system to translate sign language into speech
The technology converts South African Sign Language into spoken words in real time
A University of KwaZulu-Natal (UKZN) graduate has developed an artificial intelligence (AI) system that translates South African Sign Language (SASL) into speech in real time. The technology was created by Akhil Hansrajh as his final-year design project while completing a BSc Engineering degree in Computer Engineering, according to IOL.
The SASL-to-Speech Translation System uses computer vision, a custom AI model and a machine learning framework to track body landmarks and recognise signs. Trained on a dataset specifically developed for SASL, the system can identify individual letters and words before converting them into spoken language through a text-to-speech model.
According to the project's supervisor, Sulaiman Patel, the initiative was among the first assistive technology projects in UKZN's Electrical, Electronic and Computer Engineering discipline to focus specifically on South African Sign Language. SASL was recognised as South Africa's 12th official language, while many existing sign-language technologies internationally are designed around American Sign Language.
The project was also inspired by Hansrajh's family experience. His parents were born hard of hearing, and communication in everyday settings has often required assistance from family members. The graduate said that further development of the technology could help people who use SASL communicate more independently with those who do not know the language.
Akhil Hansrajh plans to continue improving the system's ability to recognise words and generate more cohesive and fluent speech while pursuing a career in software, networking or artificial intelligence and machine learning.
The project demonstrates how artificial intelligence and computer vision can be applied to develop technologies tailored to local languages and communication needs.