Statistical speech recognition systems typically utilise a set of statistical models of subword units based on the set of phonemes in a target language. However, in continuous speech it is important to consider co-articulation ...