Method and apparatus for generating an a priori advisor for a speech recognition dictionary

• US 5,995,929 A
• Filed: 09/12/1997
• Issued: 11/30/1999
• Est. Priority Date: 09/12/1997
• Status: Expired due to Term

• Current Assignee: Volt Delta Resources LLC (NewNet Communication Technologies LLC)
• Original Assignee: Nortel Networks Corporation
• Inventors: Gupta, Vishwa
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): Lerner, Martin
• Application Number:
  US08/928,733
• Time in Patent Office: 809 Days
• Field of Search: 704/231, 704/246, 704/244, 704/250, 704/240, 704/251, 704/252, 379/88.01, 379/88.03, 379/88.04
• US Class Current: 704/251
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: H04M 1/271 controlled by voice recognition

Abstract

The invention relates to a method and apparatus for automatically generating probability tables (histograms) for a speech recognition dictionary for use in a speech recognition system from a listing containing call records. The method comprises the step of generating histograms representing the probabilities of localities being requested based on the calling pattern collected on the field by either a human operator or via an automatic process during calls. The method is particularly useful for generating an a priori advisor for a speech recognition dictionary used in an automated system for locality recognition.

Claims

1. A method for generating a speech recognition dictionary suitable for use in a speech recognition unit of a directory assistance system, said method comprising:

providing a plurality of records of call transactions, a record including a first data element indicative of a geographical locality and a second data element indicative of at least a portion of a telephone number dialed on a terminal to access a directory assistance function;

generating from the plurality of records an a priori data structure including a plurality of probability data elements, the plurality of probability data elements being derived at least in part on the basis of the second data elements;

providing a set of vocabulary items potentially recognizable on a basis of a spoken utterance, each vocabulary item being indicative of a geographical locality;

associating at least some of the vocabulary items to probability data elements from said a priori data structure;

storing the vocabulary items and the associated probability data elements on a computer readable medium capable of being processed by the speech recognition unit of the directory assistance system to perform recognition of an utterance spoken by a user and indicative of a geographical locality, the recognition of the spoken utterance being conditioned on a basis of a telephone number dialed by the user to access a directory assistance function.

2. A method as defined in claim 1, wherein each probability data element in said a priori data structure is correlated to a frequency of a predetermined first data element in a sub-set of said plurality of records of directory assistance call transactions.

3. A method as defined in claim 2, further comprising providing in said a priori data structure a plurality of indices, each index being associated with a corresponding probability data element, each vocabulary item being associated with a corresponding index, whereby allowing to establish an association between a certain vocabulary item and a certain probability data element through an intervening index associated with said certain probability data element.

4. A method as defined in claim 3, comprising generating a plurality of a priori data structures, each a priori data structure associating probability data elements to vocabulary items.

5. A method as defined in claim 4, wherein each a priori data structure corresponds to a different telephone number at which a directory assistance function can be accessed.

6. A method as defined in claim 5, comprising assigning to each a priori data structures a data structure identifier.

7. A method as defined in claim 6, wherein the data structure identifier is indicative of at least at portion of a telephone number to be dialed to access a directory assistance function.

8. A method as defined in claim 7, wherein the at least a portion of a telephone number to be dialed to access a directory assistance function includes the NPA of a telephone number to be dialed to access a directory assistance function.

9. A method as defined in claim 1, wherein records in said plurality of records of call transactions include a third data element indicative of at least a portion of a telephone number associated to the terminal at which a directory assistance function was invoked.

10. A method as defined in claim 9, wherein said a priori data structure is a data structure of a first type, the method further comprising generating from said plurality of records an a priori data structure of a second type including a plurality of probability data elements derived at least in part on the basis of the third data elements, the method further comprising associating at least some of the vocabulary items to probability data elements from said a priori data structure of a second type.

11. A method as defined in claim 1, comprising the step of computing at least one of said probability data elements by utilizing a Turing estimate algorithm.

12. An apparatus for generating a speech recognition dictionary suitable for use in a speech recognition unit of a directory assistance system, said apparatus comprising:

a memory for holding a plurality of records of call transactions, a record including a first data element indicative of a geographical locality and a second data element indicative of at least a portion of a telephone number dialed on a terminal to access a directory assistance function;

a processor in operative relationship with said memory;

a program element suitable to be executed on said processor, said program element being operative for directing said processor to;

a) generate from said plurality of records an a priori data structure including a plurality of probability data elements, the plurality of probability data elements being derived at least in part on the basis of the second data elements;

b) map vocabulary items from a set of vocabulary items to probability data elements in said a priori data structure, the vocabulary items in the set of vocabulary items being potentially recognizable on a basis of a spoken utterance, each vocabulary item in the set of vocabulary items being indicative of a geographical locality.

13. An apparatus as defined in claim 12, wherein each probability data element in said a priori data structure is correlated to a frequency of a predetermined first data elements in a sub-set of said plurality of records of directory assistance call transactions.

14. An apparatus as defined in claim 13, wherein said a priori data structure includes a plurality of indices, each index being associated with a corresponding probability data element, each vocabulary item being associated with a corresponding index, whereby allowing to establish an association between a certain vocabulary item and a certain probability data element through an intervening index associated with said certain probability data element.

15. An apparatus as defined in claim 14, wherein said program element directs said processor to generate a plurality of a priori data structures, each a priori data structure associating probability data elements to vocabulary items.

16. An apparatus as defined in claim 15, wherein each a priori data structure corresponds to a different telephone number at which a directory assistance function can be accessed.

17. An apparatus as defined in claim 16, wherein said program element directs said processor to assign to each a priori data structure a data structure identifier.

18. An apparatus as defined in claim 17, wherein the data structure identifier is indicative of at least of portion of a telephone number to be dialed to access a directory assistance function.

19. An apparatus as defined in claim 18, wherein the at least a portion of a telephone number to be dialed to access a directory assistance function includes the NPA of a telephone number to be dialed to access a directory assistance function.

20. An apparatus as defined in claim 12, wherein records in said plurality of records of call transactions include a third data element indicative of at least a portion of a telephone number associated to the terminal at which a directory assistance function was invoked.

21. An apparatus as defined in claim 20, wherein said a priori data structure is a data structure of a first type, wherein said program element directs said processor to generate from said plurality of records an a prior data structure of a second type including a plurality of probability data elements derived at least in part on the basis of the third data elements, said program element further directing said processor to associate at least some of the vocabulary items to probability data elements from said a priori data structure of a second type.

22. An apparatus as defined in claim 21, wherein said program element directs said processor to compute at least one of said probability data elements by utilizing a Turing estimate algorithm.

23. An apparatus as defined in claim 12, wherein said program element directs said processor to compute at least one of said probability data elements by utilizing a Turing estimate algorithm.

24. A machine readable medium containing a program element for instructing a computer to generate a speech recognition vocabulary suitable for use in a speech recognition unit of a directory assistance system, said computer including:

memory means for holding a plurality of records of call transactions, a record including a first data element indicative of a geographical locality and a second data element indicative of at least a portion of a telephone number dialed on a terminal to access a directory assistance function;

a processor in operative relationship with said memory means;

a program element suitable to be executed on said processor, said program element providing means for directing said processor to;

a) generate from said plurality of records an a priori data structure including a plurality of probability data elements, the plurality of probability data elements being derived at least in part on the basis of the second data elements;

b) map vocabulary items from a set of vocabulary items to probability data elements in said a priori data structure, the vocabulary items in the set of vocabulary items being potentially recognizable on a basis of a spoken utterance, each vocabulary item in the set of vocabulary items being indicative of a geographical locality.

25. A method for generating an a priori data structure suitable for use in a speech recognition unit of a directory assistance system, said method comprising:

a) recording a multitude of directory assistance call transactions occurring in a certain geographical zone subdivided in a plurality of localities;

b) storing for each recorded directory assistance call transaction a record including a first data element indicative of a locality identified by the user during the directory assistance call transaction and a second data element indicative of at least a portion of a telephone number dialed by the user to initiate the directory assistance call transaction;

c) processing the records created at step b) to generate an a priori data structure including a plurality of probability data elements, said plurality of probability data elements being derived at least in part on the basis of said second data elements;

d) providing a set of vocabulary items, each vocabulary items being potentially recognizable on a basis of a spoken utterance;

e) associating vocabulary items form said set of vocabulary items to probability data elements of said a priori data structure, whereby allowing utilization of the probability data elements during selection of vocabulary item from said set of vocabulary items as a potential match to a spoken utterance by a user.

26. A speech recognition unit for use in a directory assistance service, said speech recognition unit comprising:

a) a first input for receiving a first signal derived from a spoken utterance indicative of a geographical locality;

b) a second input for receiving a second signal indicative of at least a portion of a telephone number dialed by a user to access a directory assistance function;

c) a processing unit coupled to said first and to said second inputs for performing speech recognition on the first signal, the speech recognition being conditioned on the second signal;

d) an output coupled to said processing unit for releasing a signal representative of a vocabulary item identified by said processing unit as being a match to the spoken utterance.

27. A speech recognition unit as defined in claim 26, comprising a speech recognition dictionary including a set of vocabulary items, the vocabulary items in the set being indicative of geographical localities and being potentially recognizable on a basis of the first signal, said processing unit being operative during the speech recognition on the first signal for determining a degree of likelihood between individual vocabulary items and the first signal.

28. A speech recognition unit as defined in claim 27, wherein said processing unit being operative to weigh on a basis of the second signal vocabulary items in the set when determining a degree of likelihood between individual vocabulary items and the first signal.