Method for determining the pronunciation variants of a word from a predeterminable vocabulary of a speech recognition system

摘要

A method for ascertaining pronunciation variations in which at least one signal sequence representing the pronunciation variations of a word are divided into segments, and by assigning at least one hidden-Markov-model- (HMM) state to one segment and on the basis of spacing intervals between the discrete signal state sequences of the word, a division of the signal state sequences into groups follows, and for each group at least one representative entry is ascertained. An independent claim is included for a computer program product.