My Ph.D. research project focused on the study of sequential Monte Carlo methods as a new tool for carrying out various signal processing tasks. More specifically, I was working on the problem of acoustic source localisation and tracking in reverberant environments, using array signal processing principles. In this specific case, particle filters offer an efficient way of dealing with the complex problem of reverberation. The project involved primarily the development and test of several particle filter algorithms using recordings of real audio data.
As a major result of this research, a particle filter using the principle of importance sampling was designed and successfully tested. This algorithm allows the automatic detection of new speakers entering the acoustic scene and also automatically recovers following the occurrence a track loss (for instance as a result of a pause in the speech signal). This specific particle filter was also implemented in real-time on a standard Linux-based desktop computer and tested in a typical office room equipped with a 16 sensor array. Some tracking examples from this algorithm have been recorded (originally running in real-time) and can be found for demonstration purposes here.
Another important part of the research was the real-time implementation of a particle-filter-based source localisation algorithm on a standard personal computer running Linux. A first version of the algorithm runs in real-time in a typical office room equipped with a 16 sensor array.
This project was undertaken under the supervision of Prof. Robert Williamson at ANU, Canberra. Part of this research was also done in joint collaboration with Dr. Darren Ward, Imperial College, London.
Other Ph.D. research interests
My Ph.D. led me to focus principally on Bayesian estimation and Sequential Monte Carlo methods, acoustic source localisation and tracking, and array signal processing. More generally however, my main research interests also include (in random order):