Joint recognition and segmentation using phonetically derived features and a hybrid phoneme model

This paper encompasses the approaches of segmental modelling and the use of dynamic features in addressing the constraints of the IID assumption in standard HMM. Phonetic features are introduced which capture the transitional dynamics across a phoneme unit via a DCT transformation of a variable length segment. Alongside this, the use of a hybrid phoneme model is proposed. Classification experiments demonstrate the potential of these features and this model to match the performance of standard HMM. The extension of these features to full recognition is explored and details of a novel recognition framework presented alongside preliminary results. Lattice rescoring based on these models and features is also explored. This reduces the set of segmentations considered allowing a more detailed exploration of the nature of the model and features and the challenges in using the proposed recognition strategy.