Speech is the type of signal humans prefer to use when directly interacting with one another. Consequently, our interaction with computers and other machines can be significantly enhanced if they, too, are able to interpret and generate natural speech. The division’s research is focused on developing signal-processing and pattern-recognition techniques to apply in speech (and also image) processing. Applications include speech recognition, biometrics and telecommunications. The research team makes use of general statistical machine learning techniques. Software development is done mainly in C++ and is deeply rooted in object-oriented design principles.
The division frequently consults for industry. Current and recently completed projects include:

• The African Speech Technology DACTS innovation fund project
• The Naval Interactive Reference Dictionary
• Afrilator: An English-Afrikaans Speech Translator
• English-isiXhosa Speech Translation

If you were to walk into the DSP Laboratory now, you would be likely to see people working on related topics, such as speech synthesis, text-to-speech technology, speech recognition, automatic speech translation, speaker identification and verification, speech search tools, and even ways of transforming a speaker’s natural voice into a new one.

We have distinguished ourselves internationally in Speech Processing: the American National Institute for Standards and Technology (NIST) hosts various international evaluations/competitions each year to measure the state of the art in various fields. These evaluations draw entries from strong research groups from some of the best known universities and industries worldwide. Since 2002 the Signal Processing Division has taken part in several of these evaluations, primarily in the field of speaker recognition, but also in language identification and spoken topic recognition. We have consistently done well in these evaluations.

Prof. TR Niesler is actively involved in the research.

Signals need not be only voltages or sounds that change over time – they can also be colours and intensities that change across two- or three-dimensional space. Therefore, the Signals Division also has an interest in processing real-world images, such as handwritten signatures and characters, or photographs and videos of faces and movement. This work is done in collaboration with the Applied Mathematics Research Group.

Prof. JA du Preez and Prof. BM Herbst (Department of Mathematical Sciences) are actively involved in the research.

Space-based communication systems are particularly interesting and challenging, since they need to be exceptionally robust, operate reliably on low power and a high-latency link, and cope with Doppler shift and radiation. The Signal Processing Division was involved in two payloads on the Sumbandila Satellite, and is already actively working on subsystems for future South African satellites. Research in this environment gives postgraduate students the opportunity to immerse themselves in the development of a complete communication system – from conceptualisation and budgeting, through design and implementation, up to testing and commissioning a functional space-based communication system. The projects also tend to be collaborative in nature, with researchers working shoulder-to-shoulder with experienced engineers in academia and industry.

Dr R Wolhuter is actively involved in the research.

Research partners include SunSpace and the Katholieke Universiteit Leuven.

Getting information from point A to point B is one thing; reliably routing that information from A through B, C, D to Z is a completely different kettle of fish. In this area, the Signal Processing Division studies protocols for data networking, teletraffic modelling, and network topology design. Here the emphasis lies very much on the telecommunications system as a whole, and on how to integrate different technologies into a coherent, reliable network.

Many of the Signal Processing Division’s research activities in this field involve improving the performance and reliability of ad-hoc networks and wireless mesh networks. These research projects also tend to be strongly application-driven, such as the deployment of wireless sensor networks for environmental monitoring, managing urban traffic on roads, or developing improved techniques for image streaming over mobile networks.

Work on Network Applications also forms part of the Telkom Centre of Excellence at Stellenbosch University.

Dr R Wolhuter is actively involved in the research.

Research partners include Telkom and Armscor.

The MIH Media Lab at Stellenbosch University was founded with the purpose to promote research in “new media” technology in South Africa. In close partnership with MIH, the internet division of Naspers, the MIH Media Lab aims to participate in research on next-generation technologies that will influence the ways in which humans interact with computers, the Web and other forms of electronic media. Dr Herman Kamper and Prof HA Engelbrecht is actively involved in the research.

The Signal Processing Division has a formal exchange agreement with the Ecole Nationale Supérieure d’Électronique, Informatique et de Radiocommunications (ENSEIRB) in Bordeaux, France, and students in or beyond their third year of engineering studies at this institution (5th-year higher education) are invited to do their internships in our laboratories. Final-year students at Stellenbosch University are also encouraged to contact us about the possibility of visiting ENSEIRB to work on their final-year projects.

Stellenbosch’s DSP Laboratory regularly hosts international students for postgraduate studies, sometimes for a short research stint, and sometimes for a full Master’s degree. You are welcome to contact us for more information.