I am a research associate in automatic speech recognition at the Centre for Speech Technology Research, in the School of Informatics
at the University of Edinburgh, where I recently completed my PhD. I am also a member of the department of
Linguistics and English Language.
Please see the CSTR pages for my
contact details
and publications.
My curriculum vitae is also available.
I came to Edinburgh after completing the MPhil in Computer Speech, Text and Internet Technology at Cambridge University, where I took my first degree, in Mathematics, a few years earlier.
I am currently working to integrating speech recognition capability into the Beetle system for tutoring in basic electronics, funded by the Office for Naval Research. This domain is an interesting one for speech recognition: since research has shown that learning achievement is highly correlated with the amount of contentful speech from the learner, the system must be designed to elict, and recognise, unstructured, large vocabulary speech, in contrast to typical spoken dialogue systems where small, prescribed grammars are often used. However, the task is different to other large vocabulary recognition tasks in that there exist tight semantic constraints.
I have been working to create the first large-vocabulary speech recogiser for Scottish Gaelic, funded by iDEA lab. The scarcity of resources for this language makes the task challenging, and is a good testing ground for recent advances in cross-lingual speech recognition.
For this project we have collected a 6-hour corpus of spoken gaelic from BBC Radio nan Gàidheal's Aithris na Maidne, fully transcribed at utterance level to modern digital standards. We may be able to make the corpus available to interested researchers in future.
For my PhD thesis I investigated the use of full covariance gaussian models for speech recognition. Full covariance models have hugely increased modelling power compared to the standard diagonal covariance model, but suffer from a number of deficiences when the quantity of training data is limited. The problem is essentially one of generalisation. I investigated two solutions: imposing sparse Gaussian Graphical Model structure on the covariance matrices by using l1-norm penalised likelihood maximisation; and by the use of a "shrinkage estimator".
My PhD supervisor was Prof. Simon King.