Colloquium Computer Science, Dr. J. Quinn

Title: Automated Radio Analysis for Disaster Response and Early Warning

                           

Abstract:

Radio is an important medium of public communication in developing countries; over 7 million words are spoken per day on the radio in Uganda alone. The high penetration of mobile phones enables members of the public to call in and air their needs, perceptions and concerns, e.g. during agricultural shocks or natural disaster. We have developed speech-to-text systems for three Ugandan languages, and used this to convert radio broadcast into text streams. Using a combination of machine learning and NLP techniques, we have been able to analyse in real time discussion pertinent to humanitarian and development planning. I will describe how this system was developed to turn unstructured radio broadcasts into a structured data source with actionable insight for the UN system.

Bio: John Quinn is a Data Scientist for United Nations Global Pulse. He is also part of the Artificial Intelligence Research group at Makerere University and currently a sabbatical visitor at the School of Informatics, University of Edinburgh. His interests are in the use of machine learning, computational statistics and artificial intelligence to solve practical problems and fill information gaps in the developing world. Homepage: http://air.ug/~jquinn