Method, system, and computer readable medium for comparing phonetic similarity of return words to resolve ambiguities during voice recognition

• US 9,318,105 B1
• Filed: 06/26/2013
• Issued: 04/19/2016
• Est. Priority Date: 10/06/2004
• Status: Active Grant

• Current Assignee: Intrado Interactive Services Corp.
• Original Assignee: West Interactive Corporation (West Corporation)
• Inventors: Khosla, Ashok M.
• Primary Examiner(s): Pullias, Jesse
• Application Number:
  US13/927,849
• Time in Patent Office: 1,028 Days
• Field of Search: 704231-257, 704270-275
• US Class Current: 1/1
• CPC Class Codes: G10L 15/06 Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice G10L15/14 takes precedence
• CPC Class Codes: G10L 15/193 Formal grammars, e.g. finite state automata, context free grammars or word networks
• CPC Class Codes: G10L 2015/221 Announcement of recognition results

Abstract

In one embodiment, the invention provides a method for a speech recognition system to select a return value corresponding to a spoken input. The method comprises generating a dictionary comprising return values associated with data provisioned in the speech recognition system; generating a grammar for each return value in the dictionary; analyzing the grammar to determine a subset of return values from the dictionary that are likely alternatives for each return value in the dictionary, based on the grammar; selecting a return value corresponding to the spoken input based on the grammar; and if the selected return value is not confirmed by a user, then presenting the likely alternative for the selected return value to the user.

Claims

1. A method, comprising:

generating a dictionary comprising return values;

generating a grammar for each return value in the dictionary;

for each return value in the dictionary, analyzing the grammar to determine a subset of return values from the dictionary that are likely alternatives for the return value, comprising for each string in the grammar for the return value, comparing each string with every other string in the dictionary that is not in the grammar for that return value;

when a first string in a pair of strings is from the grammar for a first return value, and a second string in the pair of strings is from the grammar for a second return value and is not in the grammar for the first return value, then selecting the second string as a likely alternative for the first return value, and constructing the subset of return values that are likely alternatives for each return value in the dictionary;

if said comparison indicates that the strings are related based on one of a phonetic similarity threshold and a synonym relationship then adding the return value associated with the other string to the subset; and

if the return value is not confirmed by a user, then presenting the return values in the subset for the return value to the user for selection, wherein the user is notified of each string that have a likelihood of being confused so that the user can make changes to the grammar.

2. The method of claim 1 comprising selecting the first return value corresponding to a spoken input based on the grammar.

3. The method of claim 1 comprising selecting a first return value based on the grammar.

4. The method of claim 1, wherein comparing each string uses a dynamic programming algorithm.

5. The method of claim 4, wherein the dynamic programming algorithm is based on the Smith-Waterman/Needleman-Wunsch algorithm.

6. The method of claim 1, wherein the return values are associated with data provisioned in a speech recognition system.

7. The method of claim 1, wherein the generating steps are performed at compile time as opposed to at runtime.

8. A system, comprising:

a memory configured to store logic instructions; and

a processor configured to execute the logic instructions that when executed cause the logic instructions to;

select a return value corresponding to a spoken input;

generate a subset of alternative return values for the spoken input, wherein each alternative return value is related to the selected return value based on one of a synonym relationship and a phonetic similarity threshold between a grammar for the return value and the alternative return value;

generate a first string in a pair of strings is from the grammar for a first return value, and a second string in the pair of strings is from the grammar for a second return value and is not in the grammar for the first return value, then selecting the second string as a likely alternative for the first return value, and constructing the subset of return values that are likely alternatives for each return value in a dictionary; and

present the alternative return values to a user for selection, wherein the user is notified of each string that has a likelihood of being confused.

9. The system of claim 8 wherein the logic instructions confirm the selected return value with the user.

10. The system of claim 8, wherein the user can make changes to the grammar when each string has a likelihood of being confused.

11. The system of claim 8, wherein the phonetic similarity threshold is calculated by comparing a phonetic representation of strings in the grammars for the return value and the alternative return value.

12. The system of claim 11, wherein the phonetic similarity threshold is calculated using a dynamic programming algorithm.

13. The system of claim 12, wherein the dynamic programming algorithm is based on the Smith-Waterman/Needleman-Wunsch algorithm.

14. The system of claim 8, wherein the logic to generate the subset of alternative return values generates the return values at compile time as opposed to at runtime.

15. The system of claim 8, further comprising logic to generate a grammar for each return value provisioned in the system.

16. A non-transitory computer-readable medium, having stored thereon a sequence of instructions, which when executed by a computer processor, cause the computer processor to perform:

generating a dictionary comprising return values;

generating a grammar for each return value in the dictionary;

for each return value in the dictionary, analyzing the grammar to determine a subset of return values from the dictionary that are likely alternatives for the return value comprising for each string in the grammar for the return value, comparing each string with every other string in the dictionary that is not in the grammar for that return value and if said comparison indicates that the strings are related based on one of a phonetic similarity threshold and a synonym relationship then adding the return value associated with the other string to the subset;

when a first string in a pair of strings is from the grammar for a first return value, and a second string in the pair of strings is from the grammar for a second return value and is not in the grammar for the first return value, then selecting the second string as a likely alternative for the first return value, and constructing the subset of return values that are likely alternatives for each return value in the dictionary; and

if the return value is not confirmed by a user, then presenting the return values in the subset for the first return value at once to the user for selection, wherein the user is notified of each string that has a likelihood of being confused so that the user can make changes to the grammar.

17. A system, comprising:

a processor; and

a memory coupled to the processor, the memory storing instructions which when executed by the processor, cause the computer processor to perform;

generating a dictionary comprising return values,

generating a grammar for each return value in the dictionary;

for each return value in the dictionary, analyzing the grammar to determine a subset of return values from the dictionary that are likely alternatives for the return value, comprising for each string in the grammar for the return value, comparing each string with every other string in the dictionary that is not in the grammar for that return value and if said comparison indicates that the strings are related based on one of a phonetic similarity threshold and a synonym relationship then adding the return value associated with the other string to the subset,

when a first string in a pair of strings is from the grammar for a first return value, and a second string in the pair of strings is from the grammar for a second return value and is not in the grammar for the first return value, then selecting the second string as a likely alternative for the first return value, and constructing the subset of return values that are likely alternatives for each return value in the dictionary, and

if the return value is not confirmed by a user, then presenting the return values in the subset for the first return value at once to the user for selection, wherein the user is notified of each string that has a high likelihood of being confused so that the user can make changes to the grammar.

18. A non-transitory computer-readable storage medium, having stored thereon, a sequence of instructions which when executed by a computer processor, cause the computer processor to perform:

selecting a return value corresponding to a spoken input;

generating a subset of alternative return values for the spoken input, wherein the alternative return value is related to the selected return value based on one of a synonym relationship and a phonetic similarity threshold return value based on one of a synonym relationship and a phonetic similarity threshold between the grammars for the return value and the alternative return value;

generating a first string in a pair of strings is from the grammar for a first return value, and a second string in the pair of strings is from the grammar for a second return value and is not in the grammar for the first return value, then selecting the second string as a likely alternative for the first return value, and constructing the subset of return values that are likely alternatives for each return value in a dictionary; and

presenting the alternative return values to the user for selection, wherein the user is notified of a each string that has a likelihood of being confused so that the user can make changes to the grammar.

19. A method, comprising:

selecting a return value corresponding to a spoken input;

generating a subset of alternative return values for the spoken input, wherein the alternative return value is related to the selected return value based on one of a synonym relationship and a phonetic similarity threshold between the grammars for the return value and the alternative return value;

generating a first string in a pair of strings is from the grammar for a first return value, and a second string in the pair of strings is from the grammar for a second return value and is not in the grammar for the first return value, then selecting the second string as a likely alternative for the first return value, and constructing the subset of return values that are likely alternatives for each return value in a dictionary; and

presenting the alternative return values to the user for selection, wherein the user is notified of each string that has a likelihood of being confused so that the user can make changes to the grammar.

20. The method of claim 19, wherein the step of generating the subset of alternative return values is performed at compile time as opposed to at runtime.