Systems and methods for performing speech recognition using constraint based processing

• US 8,392,193 B2
• Filed: 06/01/2004
• Issued: 03/05/2013
• Est. Priority Date: 06/01/2004
• Status: Active Grant

• Current Assignee: Rakuten Group, Inc.
• Original Assignee: Verizon Business Global LLC (Verizon Communications Inc.)
• Inventors: Schultz, Paul T., Sartini, Robert A.
• Primary Examiner(s): Godbold, Douglas
• Application Number:
  US10/857,446
• Publication Number:
  US 20050267754A1
• Time in Patent Office: 3,199 Days
• Field of Search: 704/231, 704/236, 704/252, 704/275, 379/88.04
• US Class Current: 704/270
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates

Abstract

A method for performing speech recognition includes receiving a voice input and generating at least one possible result corresponding to the voice input. The method may also include calculating a value for the speech recognition result and comparing the calculated value to a particular portion of the speech recognition result. The method may further include retrieving information based on one or more factors associated with the voice input and using the retrieved information to determine a likelihood that the speech recognition result is correct.

Claims

1. A method for performing speech recognition, comprising:

storing, by a processor, a plurality of rules associated with performing speech recognition, at least some of the plurality of rules being based on constraints associated with a subject matter of a voice input;

receiving, by the processor, a telephone call from a caller;

determining, by the processor, a telephone number from which the telephone call was made;

retrieving, by the processor and from a first database, information associated with the telephone number;

receiving, by the processor, a first voice input from the caller, the first voice input including information regarding a model year of a vehicle;

generating, by the processor, a first speech recognition result corresponding to the first voice input;

identifying, by the processor, a first group of the plurality of rules based on the first speech recognition result;

receiving, by the processor, a second voice input from the caller, the second voice input including a plurality of characters in a sequence, the plurality of characters corresponding to a first vehicle identification number;

generating, by the processor, a second speech recognition result based on the second voice input using the first group of rules, at least a first one of the first group of rules being associated with identifying one or more of the characters in the sequence and being constrained to identify one of a subset of possible characters based on a position in the sequence;

comparing, by the processor, the second speech recognition result to the retrieved information;

increasing, by the processor, a confidence value associated with the second speech recognition result when the second speech recognition result matches at least a portion of the retrieved information;

calculating, by the processor, a value using the second speech recognition result;

comparing, by the processor, the calculated value to a portion of the second speech recognition result;

increasing, by the processor, the confidence value associated with the second speech recognition result when the calculated value matches the portion of the second speech recognition result;

providing, by the processor, a first prompt to confirm a second vehicle identification number based on the second speech recognition result being associated with the increased confidence value;

receiving, by the processor, information indicating that the second vehicle identification number does not correspond to the first vehicle identification number;

applying, by the processor and based on receiving the information indicating that the second vehicle identification number does not correspond to the first vehicle identification number, a second group of the plurality of rules to the second voice input, the second group of the plurality of rules being different than the first group of the plurality of rules;

generating, by the processor, a third speech recognition result based on applying the second group of the plurality of rules to the second voice input, the third speech recognition result corresponding to a third vehicle identification number; and

providing, by the processor, a second prompt to confirm the third speech recognition result.

2. The method of claim 1, where the portion of the speech recognition result represents a predetermined character position in the second speech recognition result, the method further comprising:

eliminating the second speech recognition result as a possible speech recognition result when the calculated value does not match the portion of the second speech recognition result.

3. The method of claim 1, further comprising:

generating a plurality of speech recognition results based on the second voice input;

calculating a value for each of the plurality of speech recognition results;

comparing each calculated value to a portion of a corresponding speech recognition result, of the plurality of speech recognition results; and

determining a confidence value associated with each of the plurality of respective speech recognition results.

4. The method of claim 1, where comparing the calculated value to a portion of the second speech recognition result includes:

comparing the calculated value to a ninth character of the second speech recognition result.

5. The method of claim 1, where the generating a second speech recognition result comprises:

using the retrieved information to generate the second speech recognition result.

6. The method of claim 1, further comprising:

providing a first prompt for information to the caller; and

receiving the first voice input from the caller based on the first prompt,

where the generating a second speech recognition result includes;

using the first speech recognition result to select at least the first one of the first group of rules.

7. The method of claim 6, further comprising:

providing a second prompt for information to the caller; and

receiving the second voice input from the caller based on the second prompt,

where the generating a second speech recognition result further includes;

using the second voice input to select at least one other rule associated with performing speech recognition on the second voice input.

8. The method of claim 1, where the plurality of characters comprises a plurality of letters and numbers and the subset of possible characters includes less than all of the letters in an alphabet.

9. A system, comprising:

a memory to store a plurality of rules associated with performing speech recognition, at least a portion of the rules being based on constraints associated with a subject matter for which the speech recognition is to be performed;

an input device to receive a telephone call from a caller; and

at least one logic device to;

determine a telephone number from which the telephone call was made,

retrieve information based on the telephone number,

receive a first voice input from the caller,

generate a first speech recognition result corresponding to the first voice input,

identify a first group of the plurality of rules based on the first speech recognition result,

receive a second voice input, the second voice input including at least one of a plurality of characters in a sequence,

generate a second speech recognition result based on the second voice input using a first group of rules stored in the memory, at least a first one of the first group of rules being associated with identifying one or more characters in the sequence and being constrained to identify one of a subset of characters based on a position in the sequence, the subset of characters including less than all of a possible number of characters,

compare the retrieved information to the second speech recognition result;

provide a first prompt to confirm the second speech recognition result;

receive information indicating that the second speech recognition result does not correspond to the second voice input;

apply, based on receiving the information indicating that the second speech recognition result does not correspond to the second voice input, a second group of the plurality of rules to the second voice input, the second group of the plurality of rules being different than the first group of the plurality of rules;

generate a third speech recognition result based on applying the second group of rules to the second voice input; and

provide a second prompt to confirm the third speech recognition result.

10. The system of claim 9, where the at least one logic device is further to:

calculate a value for the second speech recognition result, and

compare the calculated value to a portion of the second speech recognition result.

11. The system of claim 10, where the at least one logic device is further to:

determine that the second speech recognition result is a potentially correct speech recognition result when the calculated value matches the portion of the second speech recognition result.

12. The system of claim 9, where the memory is further to:

store the plurality of rules based on a plurality of predetermined constraints associated with the subject matter for which the speech recognition is to be performed.

13. The system of claim 9, where the at least one logic device is further to:

provide a first prompt to the caller; and

receive the first voice input based on the first prompt,

where, when generating the second speech recognition result the at least one logic device is to;

use the first speech recognition result to select at least the first rule.

14. The system of claim 13, where the at least one logic device is further to:

provide a second prompt to the caller, and

receive second information based on the second prompt,

where when generating the second speech recognition result, the at least one logic device is further to;

use the second voice input to at select at least one other rule associated with performing speech recognition on the second voice input.

15. The system of claim 9, where when generating a second speech recognition result, the at least one logic device is to:

generate a plurality of speech recognition results, each of the plurality of speech recognition results representing a potentially correct speech recognition result.

16. The system of claim 15, where the at least one logic device is further to:

weight each of the plurality of speech recognition results, and

identify one of the plurality of speech recognition as having a highest likelihood of correctly corresponding to the second voice input based on the weighting.

17. The system of claim 9, where the plurality of characters comprises a plurality of letters and numbers and the subset of characters includes less than all of the letters in an alphabet.

18. A non-transitory computer-readable memory storing instructions, the instructions comprising:

one or more instructions which, when executed by a processor, cause the processor to determine a telephone number from which a telephone call was made;

one or more instructions which, when executed by the processor, cause the processor to access a first database to retrieve information associated with the telephone number;

one or more instructions which, when executed by the processor, cause the processor to generate a first speech recognition result corresponding to a first voice input;

one or more instructions which, when executed by the processor, cause the processor to identify a first plurality of rules based on the first speech recognition result;

one or more instructions which, when executed by the processor, cause the processor to generate a second speech recognition result based on a second voice input, the second voice input including a plurality of characters in a sequence, the one or more instructions to generate the second speech recognition result including;

one or more instructions which, when executed by the processor, cause the processor to identify at least a first one of the characters in the second voice input using at least a first one of the first plurality of rules, the first rule being associated with identifying one or more characters in the sequence based on constraints associated with a position in the sequence, the constraints defining a subset of characters for which the second speech recognition result would be correct;

one or more instructions which, when executed by the processor, cause the processor to calculate a value using at least a portion of the second speech recognition result;

one or more instructions which, when executed by the processor, cause the processor to determine whether the second speech recognition result is incorrect using the calculated value;

one or more instructions which, when executed by the processor, cause the processor to compare the second speech recognition result to the retrieved information to generate a confidence value for the second speech recognition result;

one or more instructions which, when executed by the processor, cause the processor to provide a first prompt to confirm the second speech recognition result;

one or more instructions which, when executed by the processor, cause the processor to receive information indicating that the second speech recognition result does not correspond to the second voice input;

one or more instructions which, when executed by the processor, cause the processor to apply, based on receiving the information indicating that the second speech recognition result does not correspond to the second voice input, a second one of the plurality of rules to the second voice input, the second one of the plurality of rules being different than the first one of the plurality of rules;

one or more instructions which, when executed by the processor, cause the processor to generate another speech recognition result based on applying the second one of the plurality of rules to the second voice input; and

one or more instructions which, when executed by the processor, cause the processor to provide a second prompt to confirm the other speech recognition result.

19. The computer-readable memory device of claim 18, where when determining whether the second speech recognition result is incorrect, the instructions cause the processor to:

compare the calculated value to a portion of the second speech recognition result, and

determine that the second speech recognition result is incorrect when the calculated value does not match the portion of the second speech recognition result.

20. The computer-readable memory device of claim 18, further comprising:

one or more instructions to generate a plurality of speech recognition results, each of the plurality of speech recognition results potentially corresponding to the second voice input; and

one or more instructions to weigh the plurality of speech recognition results.

21. The computer-readable memory device of claim 18, where when accessing a first database, the instructions cause the processor to:

retrieve information based on at least one of the telephone number from which the telephone call was made or a name associated with the telephone number.

22. The computer-readable memory device of claim 18, where the plurality of characters comprises a plurality of letters and numbers and the subset of characters includes less than all of the letters in an alphabet.

23. A method, comprising:

receiving a telephone call from a user;

retrieving information based on a telephone number from which the telephone call was placed;

retrieving information associated with the telephone number;

prompting the user for information;

receiving a first voice input from the user based on the prompting

generating a first speech recognition result corresponding to the first voice input, the first speech recognition result identifying a year;

identifying grammars associated with the identified year;

receiving a second voice input from the user, the second voice input including a plurality of characters in a sequence, the plurality of characters corresponding to a first vehicle identification number;

generating a second speech recognition result based on the second voice input,

the generating a second speech recognition result including;

identifying at least a first one of the characters in the second voice input using at least a first one of the identified grammars, the first grammar being associated with identifying one or more characters in the sequence based on constraints associated with a position in the sequence, the constraints defining a subset of characters for which the second speech recognition result would be correct;

providing a first prompt to confirm a second vehicle identification number based on the second speech recognition; and

receiving information indicating that the second vehicle identification number corresponds to the first vehicle identification number or does not correspond to the first vehicle identification number.

24. The method of claim 23, further comprising:

comparing the retrieved information to the second speech recognition result; and

determining a likelihood that the second speech recognition result is correct based on the comparing.

25. The method of claim 23, further comprising:

performing a mathematical operation on the second speech recognition result; and

comparing a result of the mathematical operation to a portion of the second speech recognition result.

26. The method of claim 25, further comprising:

determining that the second speech recognition result is incorrect when the result of the mathematical operation does not match the portion of the second speech recognition result.

27. The method of claim 23, further comprising:

storing a plurality of grammars associated with performing speech recognition on the second voice input, at least some of the plurality of grammars being based on constraints associated with a subject matter for which the speech recognition is to be performed and the identified year.

28. The method of claim 23, further comprising:

generating a plurality of speech recognition results;

weighting each of the plurality of speech recognition results; and

selecting one of the plurality of speech recognition results as corresponding to the second voice input based on the weighting.

29. The method of claim 23, where the plurality of characters comprises a plurality of letters and numbers and the subset of characters includes less than all of the letters in an alphabet.