System and method for targeted tuning of a speech recognition system

• US 8,401,851 B2
• Filed: 07/15/2009
• Issued: 03/19/2013
• Est. Priority Date: 08/12/2004
• Status: Active Grant

• Current Assignee: Interactions, LLC
• Original Assignee: AT&T Intellectual Property I LP (AT&T, Inc.)
• Inventors: Bushey, Robert R., Knott, Benjamin Anthony, Martin, John Mills
• Primary Examiner(s): ALBERTALLI, BRIAN LOUIS
• Application Number:
  US12/503,586
• Publication Number:
  US 20090287484A1
• Time in Patent Office: 1,343 Days
• Field of Search: None
• US Class Current: 704/244
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

A system and method of targeted tuning of a speech recognition system are disclosed. In a particular embodiment, a method includes determining a frequency of occurrence of a particular type of utterance method and includes determining whether the frequency of occurrence exceeds a threshold. The method further includes tuning a speech recognition system to improve recognition of the particular type of utterance when the frequency of occurrence of the particular type of utterance exceeds the threshold.

Claims

1. A method, comprising:

determining a frequency of occurrence of a particular type of utterance;

using a targeting engine comprising instructions that, when executed by a processor, cause the processor to determine whether the frequency of occurrence of the particular type of utterance exceeds a threshold; and

tuning a speech recognition system to improve recognition of the particular type of utterance when the frequency of occurrence of the particular type of utterance exceeds the threshold.

2. The method of claim 1, wherein tuning the speech recognition system comprises inputting a collection of utterances of the particular type of utterance to a learning module of the speech recognition system.

3. The method of claim 1, further comprising calculating a hit rate associated with the particular type of utterance.

4. The method of claim 1, further comprising calculating a false alarm value associated with the particular type of utterance.

5. The method of claim 1, further comprising maintaining a record of performance of the speech recognition system.

6. The method of claim 5, wherein an accuracy of recognition is based on the record of performance of the speech recognition system.

7. The method of claim 1, wherein an accuracy of recognition of the particular type of utterance is based on a recognition rate of the particular type of utterance.

8. A speech recognition system, comprising:

a processor;

a speech module to assign an utterance type to a received utterance;

a targeting engine comprising instructions that, when executed by the processor, cause the processor to compare a frequency of occurrence of the utterance type to a threshold; and

a tuning engine to tune the speech module to improve recognition of the utterance type when the frequency of occurrence of the utterance type exceeds the threshold.

9. The speech recognition system of claim 8, wherein tuning the speech module comprises feeding a collection of utterances of the utterance type to the speech module.

10. The speech recognition system of claim 9, wherein the collection of utterances includes one or more pre-recorded utterances stored in a repository of recordings of utterances.

11. The speech recognition system of claim 8, further comprising an accuracy engine to determine an accuracy of recognition of the utterance type.

12. The speech recognition system of claim 11, wherein the accuracy of recognition of the utterance type is compared to an accuracy threshold.

13. The speech recognition system of claim 12, wherein the tuning engine tunes the speech module to improve the accuracy of recognition of the utterance type when the accuracy of recognition of the utterance type is less than the accuracy threshold.

14. The speech recognition system of claim 8, further comprising a repository of recordings of utterances.

15. A non-transitory computer readable storage medium comprising instructions that, when executed by a processor, cause the processor to:

determine a frequency of occurrence of an utterance type;

compare the frequency of occurrence of the utterance type to a threshold; and

tune a speech recognition device to recognize the utterance type with improved accuracy when the frequency of occurrence of the utterance type exceeds the threshold.

16. The non-transitory computer readable storage medium of claim 15, further comprising instructions that, when executed by the processor, cause the processor to determine an accuracy of recognition of the utterance type.

17. The non-transitory computer readable storage medium of claim 15, further comprising instructions that, when executed by the processor, cause the processor to input a plurality of utterance recordings to the speech recognition device.

18. The non-transitory computer readable storage medium of claim 17, further comprising instructions that, when executed by the processor, cause the processor to determine a rate at which utterances of the utterance type are correctly interpreted.

19. The non-transitory computer readable storage medium of claim 17, further comprising instructions that, when executed by the processor, cause the processor to tune the speech recognition device until a desired accuracy of recognition is achieved for the utterance type.

20. The non-transitory computer readable storage medium of claim 17, further comprising instructions that, when executed by the processor, cause the processor to determine the threshold from a table of thresholds for each of a plurality of utterance types.