A method for unsupervised doping and rejection of words, in addition to the speech recognition vocabulary

摘要

The invention relates to vocal interfaces and more particularly to a vocal recognition engine, independent of the context of the particular vocal application: telephone server word recognition system, vocal dictation, on-board command and control system, recording indexing etc and a method which permits the successful use of unsupervised doping and which also noticeably improves the efficiency of the processing in vocal recognition rejection, based on a word recognition comprising a Viterbi decoding step in a Markov network to obtain a list of N candidates, associated with invalid candidate rejection steps for the doping of the application. The invention further comprises a step of doping of the application by the non-rejected candidates.