METHOD FOR DETECTING ASSOCIATIVELY PRONOUNCED WORDS

摘要

In speech recognition, speech values obtained from values sampled from a speech signal are compared with reference values, the words of a predetermined vocabulary in each case being specified by a sequence of reference values. The words are specified in terms of phonemes in accordance with a previously specified dictionary of pronunciation and the reference values for the phonemes are determined in a learning phase, each phoneme within a word consisting of a number of similar reference values, this number having been determined in the learning phase. To approximate transitions between phonemes, each phoneme can also consist of three segments each of constant reference values. The predetermined number of reference values per phoneme permits improved simulation of the time duration of a phoneme within a particular word. Various possibilities are specified for determining the reference values and the distance value in the recognition process.