Research interests
This page gives a general overview of my research interests and experience. For more detailed information about some of the specific projects I have worked on, please click on the sub-tabs above.
Human language and machine learning
Overall, my research is motivated on one hand by a long-held interest in human language, how it is produced and perceived, and on the other hand by strong enthusiasm for machine learning, whereby we train computer models using representative data to perform human-like processing instead of programming them explicitly for a task. I have pursued these two interests, as student and researcher, for more than twenty years, most often combining the two for the development of speech-related technologies, but also in other applications of machine learning. My research can be categorised broadly into four areas:
Articulatory modelling
Since my PhD studies, I have been interested in the relationship between the articulatory and acoustic domains, and have worked extensively on modelling the acoustic-articulatory (inversion) mapping, the articulatory-acoustic (synthesis) mapping, as well as on the acquisition and modelling of articulatory movement and shape data itself.
Speech synthesis and recognition
I am keenly interested in ways to exploit an articulatory(-like) representation of speech to improve speech technology, having worked with hidden Markov model and linear dynamical model systems for automatic speech recognition, as well as extensive work with state-of-the-art concatenative and statistical parametric speech synthesis.
Pronunciation modelling
I hold a significant interest in several aspects of pronunciation modelling, including: design and implementation of pronunciation lexica; machine learning for automatic letter-to-sound conversion; connected-speech processes; regional accent phonetics/phonology and pronunciation variation; pronunciation modelling from the perspective of derivational morphology.
Other applications of machine learning
Besides speech, I have been keen to tackle problems in other areas where machine learning is also of huge utility, e.g. tracking shapes in ultrasound video, validation of carbon fibre microscopy images, and trainable magnetic field modelling.