The Lexical Access Component of the CMU Continuous Speech Recognition System

The CMU Lexical Access system hypothesizes words  from a phonetic  lattice,  supplemented by a coarse labelling of the speech signal.  Word hypotheses are anchored on syllabic nuclei and  are generated  independently  for different parts of the  utterance.  Junctures  between words are resolved separately, on  demand from  the  Parser module.  The lexical representation is  generated by rule from base foms ,  in a completely  automatic process.  A description of the various components of the  system is provided, a s  well a s  performance  data.