System and method for augmenting spoken language understanding by correcting common errors in linguistic performance

• US 20050192801A1
• Filed: 02/26/2004
• Published: 09/01/2005
• Est. Priority Date: 02/26/2004
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: AT&T Corporation (AT&T, Inc.)
• Inventors: Rosen, Kenneth H., Lewis, Steven H.
• Granted Patent: US 7,505,906 B2
• Time in Patent Office: Days
• US Class Current: 704/240
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams

Abstract

A method and system for automatic speech recognition are disclosed. The method comprises receiving speech from a user, the speech including at least one speech error, increasing the probabilities of closely related words to the at least one speech error and processing the received speech using the increased probabilities. A corpora of data having common words that are mis-stated is used to identify and increase the probabilities of related words. The method applies to at least the automatic speech recognition module and the spoken language understanding module.

Claims

1. A method of processing speech, the method comprising:

receiving speech from a user, the speech including at least one speech error;

modifying probabilities of closely related words to the at least one speech error; and

processing the received speech using the increased probabilities.

2. The method of claim 1, wherein processing the received speech further comprises automatic speech recognition of the received speech using the modified probabilities.

3. The method of claim 1, wherein processing the received speech further comprises spoken language understanding of the received speech using the modified probabilities.

4. The method of claim 1, wherein the closely related words are words that begin and end with similar sounds to the at least one speech error.

5. The method of claim 1, wherein the closely related words are words that begin and end with the same phoneme and that have the same number of syllables as the at least one speech error.

6. The method of claim 1, wherein the speech error comprises a plurality of words, and the closely related words sound similar to the plurality of words.

7. The method of claim 1, further comprising using a corpora of data associated with speech errors.

8. The method of claim 1, further comprising using a corpora of data associated with the user'"'"'s language patterns.

9. The method of claim 8, wherein the user'"'"'s language patterns relate to the language spoken by the user.

10. The method of claim 9, wherein the corpora of data further comprises common speech errors made by speakers of the language spoken by the user.

11. An language processing module in a spoken dialog system, the module comprising:

means for receiving speech from a user, the speech including at least one speech error;

means for modifying the probabilities of closely related words to the at least one speech error; and

means for processing the received speech using the modified probabilities.

12. The module of claim 11, wherein the module is an automatic speech recognition module.

13. The module of claim 11, wherein the module is spoken language understanding module.

14. The module of claim 11, wherein the closely related words are words that begin and end with similar sounds to the at least one speech error.

15. The module of claim 11, wherein the speech error comprises a plurality of words, and the closely related words sound similar to the plurality of words.

16. The module of claim 11, further comprising using a corpora of data associated with speech errors.

17. The module of claim 11, further comprising using a corpora of data associated with the user'"'"'s language patterns.

18. The module of claim 11, wherein the closely related words are words that begin and end with the same phoneme and that have the same number of syllables as the at least one speech error.

19. The module of claim 17, wherein the user'"'"'s language patterns relate to the language spoken by the user.

20. The module of claim 19, wherein the corpora of data further comprises common speech errors made by speakers of the language spoken by the user.

21. A spoken dialog system having a speech processing module, the module comprising:

means for receiving speech from a user, the speech including at least one speech error;

means for modifying probabilities of closely related words to the at least one speech error; and

means for processing the received speech using the modified probabilities.

22. The spoken dialog system of claim 21, wherein the module is an automatic speech recognition module.

23. The spoken dialog system of claim 21, wherein the module is a spoken language understanding module.

24. The spoken dialog system of claim 21, wherein the closely related words are words that begin and end with similar sounds to the at least one speech error.

25. The spoken dialog system of claim 21, wherein the speech error comprises a plurality of words, and the closely related words sound similar to the plurality of words.

26. The spoken dialog system of claim 21, further comprising using a corpora of data associated with speech errors.

27. The spoken dialog system of claim 21, further comprising using a corpora of data associated with the user'"'"'s language patterns.

28. The spoken dialog system of claim 27, wherein the user'"'"'s language patterns relate to the language spoken by the user.

29. The spoken dialog system of claim 28, wherein the corpora of data further comprises common speech errors made by speakers of the language spoken by the user.

30. The spoken dialog system of claim 21, wherein the closely related words are words that begin and end with the same phoneme and that have the same number of syllables as the at least one speech error.

31. A method of performing automatic speech recognition, the method comprising:

determining whether a user'"'"'s utterances relate to one of a plurality of corpora of data having modified probabilities of certain words;

if the user'"'"'s utterances relate to one of the plurality of corpora of data, selecting the related corpora of data for use in speech recognition;

receiving speech from a user, the speech including at least one speech error; and

using the selected corpora of data having modified the probabilities to process the received speech to increase the probability of correctly interpreting the received speech with the at least one speech error.

32. The method of claim 31, wherein the words having modified probabilities in the corpora of data include words that begin and end with the same phoneme and that have the same number of syllables as the at least one speech error.

33. The method of claim 31, wherein the speech error comprises a plurality of error words, and the corpora of data relates to pluralities of words that are closely related in sound to the plurality of error words.

34. The method of claim 31, wherein using the selected corpora of data having modified the probabilities to process the received speech further comprises performing automatic speech recognition on the received speech.

35. The method of claim 31, wherein using the selected corpora of data having increased the probabilities to process the received speech further comprises performing spoken language understanding on the received speech.

36. A method of recognizing speech using a speech recognition module, the method comprising:

receiving speech from a user, the speech including at least one speech error;

determining whether the speech recognition module correctly interpreted the one speech error according to what the user intended to say;

modifying probabilities of recognition parameters according to results of the determining step; and

recognizing the received speech using the increased probabilities.

37. The method of claim 36, wherein modifying the probabilities of recognition parameters further comprises increasing the probabilities of recognition parameters.

38. The method of claim 36, wherein determining whether the speech recognition module correctly interpreted the one speech error according to what the user intended to say further comprises a confirmation prompt to the user.

39. The method of claim 36, wherein modifying probabilities of recognition parameters according to results of the determining step allows the speech recognition module to correct speech errors for the particular user speaking.