This paper presents an approach to lexical access in continuous speech recognition systems based on dynamic time warping at the phoneme level, implemented using finite-state transducers. Robustness is thereby achieved in the face of both contingent and necessary imperfections in the phoneme lattice from which lexical access starts. Issues of training, evaluation and context-sensitivity are discussed.
Bibliographic reference. Thompson, Henry S. / McKelvie, David / McInnes, Fergus R. (1989): "Robust lexical access for continuous speech using dynamic time warping and finite-state transducers", In EUROSPEECH-1989, 2059-2062.