Many rural communities across Africa rely on radio broadcasts as their primary source of information. Radio shows also provide a platform for interactive discussion of issues that matter the most to the public through call-in shows. In Uganda alone, there are approximately 250 FM radio stations, many of them operated at the community level. The audio streams transmitted via radio are therefore a valuable source of insight into people’s opinions and perceptions, as well as a reflection of how effectively official figures communicate with citizens about different issues. However, capturing relevant information from this data source is challenging; around 7.5 million words are transmitted daily, audio quality can vary drastically and there are different regional languages to work with.

In 2015, the United Nations Development Programme's Global Pulse team in Kampala developed a prototype analysis tool using distributed hardware to capture radio content data from around Uganda. They were interested in understanding public attitudes towards the Sustainable Development Goals. The team developed three natural language processing (NLP) models to convert speech to text through speech recognition and translation between English and the regional languages Luganda and Acholi. The tool automatically tracked trends in the use of terms relevant to the global goals.

This example of AI and collective intelligence demonstrated the potential of using non-conventional channels, such as radio, as a form of citizen-sensing to identify topics of relevance to members of the public. Although the project was short-lived, it provided a valuable proof of concept that citizen-generated audio contains unique information that is of use to the public and third sector. Building AI tools that are able to interpret regional languages also helps to ensure that more diverse views are incorporated into conversations about sustainable development, widening the reach of underrepresented communities.