| 摘要 |
The invention relates to a speech-identification process, in which a speech signal is received 1 and signal processing is carried out 3, whereupon one or more conversion correspondences are formed 5, on the basis of which possible identification findings 11 are produced from the vocabulary 7. In the process, knowledge of the speech structure 9 is used to eliminate 13 as well as to enable continued use 15 of the possible identification findings 11 which have emerged. Fundamental to the invention is that, as a criterion in the elimination of the identification findings as well as in the enablement of continued use, information is used which is obtained from a previously investigated sound location distribution regarding the probability of a particular sound being situated somewhere within a word. From the resultant possible identification findings, those are eliminated for which the probability value obtained for the sound location distribution is less than a given limit value. Correspondingly, the opportunity is given to continue towards those identification findings for which the probability value obtained for the sound location distribution exceeds the given limit value. <IMAGE>
|
