Prof Barnard with Prof Roelf Sandenbergh (UP), digital@SERA Project Team Leader.

Task Team Profile - digital@SERA

digital@SERA is leading the evolution in Human Language Technology (HLT) research, with group members working on projects including speech synthesis, speech recognition and machine translation. These technologies are widely deployed across developed countries, while the developing world is only now starting to realise their value.

Since its launch in 2004, digital@SERA - an IT research platform across the CSIR's Meraka Institute and its Defence, Peace, Safety & Security business unit, and the University of Pretoria's Faculty of Engineering, the Built Environment and Information Technology - has continuously focused on conducting research that is sustainable as well as commercially and socially relevant. Initial flagship programmes were identified in Human Language Technologies, Information and Computer Security and Open Source software.

Prof Etienne Barnard, the joint leader of the HLT research group is an extraordinary professor in UP's Department of Computer Science and a research scientist at Meraka. "We have always tried to maintain a dual focus in the work we do: a commitment to good science and high quality research, but also - and in a sense more importantly - to contribute to making a difference in society."

According to group co-leader Dr Marelie Davel, a research scientist at the Meraka Institute, this is one way in which the collaborative effort has been very beneficial. "Apart from increasing critical mass by joining efforts, the various Meraka projects provide an environment that can absorb research outputs quickly and in innovative ways."

The major HLT projects all aim at providing information to people who would otherwise not easily be able to access it. Currently, the biggest focus is on speech synthesis or Text to Speech (TTS) technology. The first projects concentrated on isiZulu synthesis, and according to Barnard these systems can now produce fairly understandable, although not yet natural-sounding, synthesis.

Speech recognition is a more complex task, as these systems need to recognise different languages and voices, as well as nuances such as dialects and accents. "Even the best systems in the local languages are perhaps 80% accurate, while the best English systems produce an accuracy of about 98%", says Barnard. Undaunted by the challenges, the HLT group aims to develop a local speech recognition system that works as well as English systems within the next five years.

Two systems have already been developed to a level where people are starting to use them in practical applications. The first of these is "OpenPhone" - a telephone-based system for a local NGO active in HIV/AIDS prevention and treatment. With a HLT speech synthesis system connected to the NGOs main information system, its field workers can telephonically access up-to-date information on new treatments from remote locations.

The speech synthesis systems have also been integrated Meraka's National Accessibility Portal (NAP) to enhance the independence of persons with disabilities. As part of NAP the speech systems allow people with speech disabilities to communicate via synthesised speech. The same technologies can similarly be used to enable screen readers in local languages for people with vision disabilities.

digital@SERA has always focused on facilitating collaboration and influencing researchers to direct their research in useful directions. In addition, an important goal is to extend the research pipeline. This requires significant investment, as it can take up to three years to get the pipeline to a sustainable level, but there is a lot of interest amongst team young researchers at UP in the potential of the HLT field. "Growth of this pipeline is something that we don't have enough of in South Africa, and it is very encouraging to see young people getting excited about this kind of work", according to Barnard.