Method of determining database entries
The invention relates to a method of determining database entries of a database ( 9 ) by means of an automatic dialog system ( 1 ) in which the following steps are provided: 1.1 temporary storage of a speech input, 1.2 processing of the speech utterance by means of an automatic speech recognition device ( 4 ) and generation of a speech recognition result using a first lexicon ( 6 ) which has a first set of lexicon entries, where a lexicon entry contains at least one acoustic reference, 1.3 search for a database entry corresponding to the speech recognition result, 1.4 for the case where no database entry is found in step 1.3, adaptation of the first lexicon ( 6 ) so that the first lexicon ( 6 ) has a second set of lexicon entries which differs from the first set of lexicon entries by at least one lexicon entry, repetition of steps 1.2 and 1.3 using the temporarily stored speech utterance and the adapted first lexicon ( 6 ).
1. A method of determining database entries of a database using an automatic dialog system, the method comprising:
temporarily storing speech input,
processing speech input using an automatic speech recognition device and generating a speech recognition result using a first lexicon, wherein the first lexicon comprises a first set of lexicon entries, and each lexicon entry contains at least one acoustic reference, which corresponds to a Hidden Markov Model,
searching the database for a database entry corresponding to the speech recognition result,
after the searching, and in a case where no database entry is found, adapting the first lexicon to include a second set of lexicon entries without additional user input, wherein the second set of lexicon entries differs from the first set of lexicon entries by at least one lexicon entry, and repeating the processing and the searching using the temporarily stored speech input and the adapted first lexicon; and
wherein, in a case where a database entry is found, outputting.
2. The method as claimed in claim 1 , wherein the first lexicon entries are taken from a second lexicon which serves as an overall lexicon.
3. The method as claimed in claim 1 , wherein the processing, the searching and the adapting are repeated until an interrupt criterion is fulfilled.
4. The method as claimed in claim 1 , wherein the step of adapting includes providing categories of lexicon entries, corresponding to categories of database entries, and wherein the adaptation of the first lexicon with respect to a category of lexicon entries deviates from the first lexicon in at least one lexicon entry.
5. The method as claimed in claim 1 , wherein categories of lexicon entries are provided that correspond to categories of database entries, and wherein a second category of lexicon entries is allocated, which includes at least a subset of entries of the first lexicon adaptation, said second category comprising at least a subset of a category of lexicon entries corresponding to the first lexicon.
6. The method as claimed in claim 1 , wherein a speech recognition result is supplied by a speech recognition device, said result comprising a number N of best recognition result alternatives if N>1, and wherein the adaptation of the lexicon depends on at least one recognition result alternative.
7. The method as claimed in claim 1 , wherein the database entries are parts of a Yellow Pages data collection.
8. The method as claimed in claim 1 , wherein the dialog system has a telephone interface by which speech inputs and speech outputs arc transmitted.
9. An automatic dialog system for determining database entries of a database, said system including a plurality of system components for carrying out the following steps:
temporarily storing of a speech utterance to be input,
processing of the speech utterance using an automatic speech recognition device and generating a speech recognition result using a first lexicon, said first lexicon including a first set of lexicon entries, and wherein each lexicon entry comprises at least one acoustic reference, which corresponds to a Hidden Markov Model,
searching for a database entry corresponding to the speech recognition result,
after the searching, and, in a case where no database entry is found, adapting the first lexicon to include a second set of lexicon entries without additional user input, wherein the second set of lexicon entries differs from the first set of lexicon entries by at least one lexicon entry, and repeating the processing and the searching using the temporarily stored speech utterance and the adapted first lexicon; and
wherein, in a case where the database entry is found, outputting.