Connected-word recognition system using speaker-independent phonetic features

A speaker-independent connected Word recognition system using pertinent phonetic features for French language has been simulated on a PDP 11-34. Its originality is based on the fact that the system doesn´t require any oral training ; the user has only to type the vocabulary while the system automaticaly converts the phonemic description of the vocabulary into an ideal and speaker-independent sequence of phonetic features which will be compared with the feature sequence of the unknown utterance in the recognition process. Such a procedure allows to build an ideal dictionnary and in this way it increases speaker-independent performances.