Speech recognition disambiguation of homophonic ending words

摘要

The invention is related to a method for performing electronic speech recognition comprising the steps of: performing speech recognition (56) on an utterance; generating a recognition candidate list (24) comprising at least one recognition candidate (26) from the result of the speech recognition (56) by using a directory (28) of possible candidates (26).;For making discrete speech recognition processing more robust in noisy environment it is proposed to review (50) the candidate list (24) for existence of at least one noise sensitive candidate (26) being similar to at least one directory entry upon truncation of the directory entry, the candidate (26) or both; and depending on a review result re-performing speech recognition (58) on the utterance.