A year and a half after it was prototyped, the radio content analysis tool developed by Pulse Lab Kampala and partners has become fully operational.
The findings and lessons learned during the process were compiled in a report entitled: “Using Machine Learning to Analyse Radio Content in Uganda – Opportunities for Sustainable Development and Humanitarian Action.”
The recent Artificial Intelligence (AI) for Good Global Summit has brought together partners to define a roadmap for governments, industry, academia, media, and civil society to develop AI in a safe, responsible and ethical manner benefiting all segments of society.
At the summit, the radio content analysis tool was showcased as one of the applications of AI currently in use at the UN.
The tool was designed to leverage public radio content as a source of information to inform on issues relevant to sustainable development. The most complex part in the development of the prototype is capturing the transcription of spoken words into written text.
This technology, called speech recognition, is used in applications ranging from simple voice dialing (e.g. “Call home”) to fully automatic speech-to-text processing where every word is being converted into text (e.g. dictation to a document or email).
The world’s largest IT companies, including Apple, Google, Microsoft and IBM, invest significant resources in speech recognition for their products. There are also companies that specialise in speech recognition as Nuance Communications (Apple’s supplier) or HTK.
This type of companies offer automatic speech-to-text dictation in about 50 languages, but languages and dialects from the African continent are not available among them.
The radio content analysis tool was developed as part of a project conducted by Pulse Lab Kampala in collaboration with the Stellenbosch University in South Africa. The tool works by converting public discussions that take place on radio in various African languages into text.
Once converted, the text can be searched for topics of interest. The tool is now fully functional in the Northern and Central regions of Uganda and available for three languages: Luganda, Acholi and English (as spoken in the country).
The report outlines the methodology and processes of the radio content analysis tool, distills the technology behind its creation and presents the lessons learned along the way.
It also details the results of several pilot studies that were conducted together with partners from the Government, UN agencies and academia to understand the validity and value of unfiltered public radio discussions for development.
The hope is that the processes and lessons detailed in the report can serve as examples and inspiration for using radio talk and data analytics to inform decision-making processes in development and humanitarian scenarios, in contexts where other sources of data may be missing or insufficient.