Development and
Cooperation
Many countries in Africa lack qualified sign language interpreters. This makes it difficult for deaf and hearing-impaired people to access education, employment and public services. This is where Terp 360 comes in, an AI-powered application developed by Kenyan innovator Elly Savatia. The idea for Terp 360 was born during a class in Northern Kenya, where Savatia noticed a critical issue: there was only one interpreter for a total of 300 deaf students.
The app translates spoken language into sign language and displays the translation using 3D avatars. These avatars are designed not only to string individual signs together, but to convey them in a natural-looking form.
Terp 360 is based on a growing dataset of more than 2300 locally recorded signs that are intended to establish a culturally relevant connection. Deaf and hearing-impaired people were involved in the development of the app, which won the Africa Prize for Engineering Innovation last year.
Terp 360 has already interacted with more than 2000 members of the deaf and hearing-impaired community. Savatia plans to expand the app’s vocabulary, dialects and colloquialisms.
Artificial intelligence quickly reaches its limits when it comes to languages for which there is little training data. Two models from Africa and India are addressing precisely this issue.
The company Lelapa AI is developing InkubaLM, a multilingual large language model (LLM) that can understand and generate text. InkubaLM focuses on African languages that are digitally underrepresented. At launch, the model supports Kiswahili, Yoruba, isiXhosa, Hausa and isiZulu. It is designed to enable basic language tasks such as translation and transcription, as well as other automatic language processing procedures. InkubaLM is based on two data sets. Lelapa AI provides the model and resources as open source.
Sarvam AI, founded in 2023 by Vivek Raghavan and Pratyush Kumar, works with datasets for 22 official Indian languages and is developing several components: a model for automatic speech recognition for 10 Indian languages, a translation model for 110 language pairs, and a speech synthesis model that can read documents, including historical and multilingual ones.
In parts of Africa, there is a particular shortage of specialists in diagnostic imaging. Radiologists evaluate X-ray, CT or mammography images and identify signs of disease. Where there are too few of these specialists, diagnoses are delayed – with consequences for treatment and survival rates. This is the starting point for the Ghanaian company MinoHealth AI Labs, founded by AI expert Darlington Akogo.
The approach: AI systems should automatically evaluate medical images and thus support medical teams – cost-effectively and in less than a minute. According to its own information, MinoHealth AI is developing models that can detect 14 findings on chest X-rays – for example, pneumonia, fibrosis or fluid between the lungs and chest wall. The AI is also intended to help detect breast cancer in mammograms. A second field is infectious diseases. MinoHealth AI is developing AI systems for malaria, Covid-19 screening and tuberculosis-related damage that can be seen in X-rays.
Such applications do not replace clinical expertise, but they can help to make more targeted use of scarce resources – provided that data quality, clinical validation and responsible use are in place.
The D+C editorial team
euz.editor@dandc.eu
Older populations
Inclusion
Urban mobility
AI tools
Digitalisation
West Africa
Good News
Politics
International law
Disability rights
Gender equality