Radio is a vibrant medium for public discussion in Uganda. Talk shows and phone-ins hosted by community radio stations are popular ways for Ugandans to voice their needs, concerns and opinions—particularly in rural areas and for older people. We are developing an innovative technology to test the use of this for early warning about issues affecting development.
In Uganda, radio is one of the main ways people from vulnerable groups can publically express themselves. Community radio stations in particular—often operating in modest circumstances—are used, for example, to discuss issues in health, governance, gender issues and agriculture. Although radio is often thought of as a one-way communication, its popularity in Uganda can be attributed partly to the interactive way in which it is used; the high penetration rate of mobile phones means that it is now easy for listeners to call or text their feedback.
Community radio in Nakaseke (image: UNESCO)
In 2014 there were 216 registered FM radio stations across the country, broadcasting on 299 different transmitters (source: Uganda Communications Commission). The following map shows their distribution across the country:
Distribution of FM transmitters in Uganda, plotted using information provided to Global Pulse by the Uganda Commmunications Commission. Circle sizes denote the number of transmitters in a given district.
This project builds on recent work using the monitoring of public social media communication as an innovative method for early warning and the monitoring and evaluation of development projects. We propose to use public radio discussion in a similar way.
In order to obtain such insights from radio discussion, the system we are working towards is envisaged as follows.
The technology to be developed in order for this to work breaks down into three main parts: (1) the means to physically receive and archive radio broadcasts from around the country, (2) the software needed to provide limited interpretation of speech signals, and (3) the analytics to derive insight from this data.
To record radio broadcasts, we are currently using RTL-SDR hardware and open source GNU Radio software, which allows us to receive multiple radio stations with a single device. We are currently looking at two configurations: a desktop/laptop with several RTL-SDR receivers in order to collect all stations in urban areas, and a mobile unit based on the Raspberry Pi, a credit card sized computer, for easy deployment in rural areas.
RTL-SDR receiver with Raspberry Pi.
Multiple FM radio stations in Kampala picked up by a single device.
Developing the software needed to give some interpretation of recorded speech signals is technically the most complex part of the project. Although we are looking at a constrained form of speech-to-text technology known as word spotting, we have to do this with noisy and lossy audio channels (e.g. phone-in calls, with background noise, broadcast on FM), with usually informal styles of speech and multiple speakers, and in Ugandan languages for which there are few existing linguistic resources.
To do this we are working with speech technology experts in Stellenbosch University to first adapt South African speech models to Ugandan pronunciation of English, and later to develop new models for vernacular languages including Acholi and Luganda. We are focussing on the open source software HTK as a platform for this.
We aim to derive two types of analysis results from the system:
UN Global Pulse has a set of privacy and data protection principles, to ensure that the individuals represented in the data we access are not adversely affected by our research. In this project, we only access radio discussions which were made for the purpose of public broadcast (i.e. where those speaking have the intention of communicating publically), and only carry out analysis on topics related to UN development and humanitarian goals.