| 摘要 |
<p>In the recognition of words from a large vocabulary the comparison of the spoken signal in each case in parallel with all the words requires an excessively large expenditure of time and the recognition reliability is low. In addition, in the case of words spoken in connected fashion the word boundaries cannot always be readily determined. In the present invention the natural grammar is taken into account to a certain extent and is given in the form of a list which contains the assignment in each case of a syntactic category to a plurality of other categories and/or words. As a result, the number of words which, during the recognition process, have to be handled in each case simultaneously and which are stored in a second list can be considerably reduced in each case. In addition, a third list with information relating to hypotheses for the sentence to be recognised is built up from the first list, the hypotheses being partially verified and further possible hypotheses for the voice signal to follow being newly constructed as the comparison of the voice signal with the current words and the result thereof progresses. …<IMAGE>… </p> |
