Error correction in speech recognition by correcting text around selected area

• US 6,912,498 B2
• Filed: 05/02/2001
• Issued: 06/28/2005
• Est. Priority Date: 05/02/2000
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: ScanSoft, Inc. n/k/a Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Stevens, Daniell, Roth, Robert, Newman, Michael J., Sturtevant, Dean, Abrahams, David, Gould, Joel M., Ingold, Charles E., Gold, Allan
• Primary Examiner(s): Chawan, Vijay B.
• Application Number:
  US09/845,769
• Publication Number:
  US 20020138265A1
• Time in Patent Office: 1,518 Days
• Field of Search: 704/235, 704/256, 704/254, 704/251, 704/244, 704/200
• US Class Current: 704/235
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 2015/0631 Creating reference templates; Clustering

Abstract

Correcting incorrect text associated with recognition errors in computer-implemented speech recognition includes receiving a selection of a word from a recognized utterance. The selection indicates a bound of a portion of the recognized utterance to be corrected. A first recognition correction is produced based on a comparison between a first alternative transcript and the recognized utterance. A second recognition correction is produced based on a comparison between a second alternative transcript and the recognized utterance. The duration of the first recognition correction differs from the duration of the second recognition correction. A portion of the recognition result that is replaced with one of the first recognition correction and the second recognition correction. includes at one bound a word indicated by the selection and extends for the duration of the one of the first recognition correction and the second recognition correction with which the portion is replaced.

Claims

1. A method of correcting incorrect text associated with recognition errors in computer-implemented speech recognition, the method comprising:

performing speech recognition on an utterance to produce a recognition result for the utterance;

receiving from a user a selection of a word from the recognized utterance, the selection indicating a bound of a portion of the recognized utterance to be corrected;

comparing a first alternative transcript to the recognized utterance to be corrected;

producing a first recognition correction based on the comparison;

comparing a second alternative transcript to the recognized utterance to be corrected;

producing a second recognition correction based on the second comparison; and

replacing a portion of the recognition result with one of the first recognition correction and the second recognition correction;

wherein a duration of the first recognition correction differs from a duration of the second recognition correction, and the portion of the recognition result replaced includes at one bound a word indicated by the selection and extends for the duration of the one of the first recognition correction and the second recognition correction with which the portion is replaced and includes at least one word other than the user-selected word.

2. The method of claim 1 wherein the selection indicates a beginning bound of a recognized utterance to be corrected.

3. The method of claim 1 wherein the selection indicates a finishing bound of a recognized utterance to be corrected.

4. A method of correcting incorrect text associated with recognition errors in computer-implemented speech recognition, the method comprising:

performing speech recognition on an utterance to produce a recognition result for the utterance;

receiving a selection of a word from the recognized utterance, the selection indicating a bound of a portion of the recognized utterance to be corrected;

comparing a first alternative transcript to the recognized utterance to be corrected;

producing a first recognition correction based on the comparison;

comparing a second alternative transcript to the recognized utterance to be corrected;

producing a second recognition correction based on the second comparison; and

replacing a portion of the recognition result with one of the first recognition correction and the second recognition correction;

wherein a duration of the first recognition correction differs from a duration of the second recognition correction, and the portion of the recognition result replaced includes at one bound a word indicated by the selection and extends for the duration of the one of the first recognition correction and the second recognition correction with which the portion is replaced, and

wherein comparing an alternative transcript to the recognized utterance comprises;

selecting from the alternative transcript a test word that is not identical to the selected word and that begins at a time that is nearest a time at which the selected word begins; and

searching in time, through the recognized utterance relative to the selected word, and through the alternative transcript relative to the test word, until finding a word common to the recognized utterance and the alternative transcript.

5. The method of claim 4 wherein the common word begins at a time in the recognized utterance that is substantially equal to a time at which the common word begins in the alternative transcript.

6. The method of claim 5 wherein producing a recognition correction comprises selecting a word string from the alternative transcript, the word string bound by the test word from the alternative transcript and by a word from the alternative transcript that is adjacent to the common word and between the test word and the common word.

7. The method of claim 6 further comprising receiving a selection of one of the first recognition correction and the second recognition correction.

8. The method of claim 6 wherein searching in time through the recognized utterance and through the alternative transcript comprises:

designating a word adjacent to the test word as an alternative transcript word;

designating a word adjacent to the selected word as an original transcript word; and

comparing the original transcript word to the alternative transcript word.

9. The method of claim 8 wherein searching in time through the recognized utterance and through the alternative transcript comprises:

designating the original transcript word and the alternative transcript word as the common word if;

the original transcript word is identical to the alternative transcript word, and

a time at which the original transcript word begins is near a time at which the alternative transcript word begins.

10. The method of claim 9 wherein searching in time through the recognized utterance and through the alternative transcript comprises:

designating a word adjacent to the alternative transcript word in the alternative transcript as the alternative transcript word if;

the original transcript word is not identical to the alternative transcript word, and

a time at which the original transcript word begins is later than a time at which the alternative transcript word begins.

11. The method of claim 9 wherein searching in time through the recognized utterance and through the alternative transcript comprises:

designating a word adjacent to the alternative transcript word in the alternative transcript as the alternative transcript word if;

the original transcript word is identical to the alternative transcript word, and

a time at which the original transcript word begins is later than a time at which the alternative transcript word begins.

12. The method of claim 9 wherein searching in time through the recognized utterance and through the alternative transcript comprises:

designating a word adjacent to the original transcript word in the original transcript as the original transcript word if;

the original transcript word is not identical to the alternative transcript word, and

a time at which the original transcript word begins is earlier than a time at which the alternative transcript word begins.

13. The method of claim 9 wherein searching in time through the recognized utterance and through the alternative transcript comprises:

designating a word adjacent to the original transcript word in the original transcript as the original transcript word if;

the original transcript word is identical to the alternative transcript word, and

a time at which the original transcript word begins is earlier than a time at which the alternative transcript word begins.

14. The method of claim 9 wherein searching in time through the recognized utterance and through the alternative transcript comprises:

designating a word adjacent to the original transcript word in the original transcript as the original transcript word and a word adjacent to the alternative transcript word in the alternative transcript as the alternative transcript word if;

the original transcript word is not identical to the alternative transcript word, and

a time at which the original transcript word begin is near a time at which the alternative transcript word begins.

15. A method of correcting incorrect text associated with recognition errors in computer-implemented speech recognition, the method comprising:

performing speech recognition on an utterance to produce a recognition result for the utterance;

receiving a selection of a word from the recognized utterance, the selection indicating a bound of a portion of the recognized utterance to be corrected;

comparing an alternative transcript to the recognized utterance to be corrected, the comparing comprising;

selecting from the alternative transcript a test word that begins at a time that is nearest a time at which the selected word begins; and

searching in time, relative to the selected word, through the recognized utterance, and searching in time, relative to the test word, through the alternative transcript until finding a word common to the recognized utterance and the alternative transcript;

producing a recognition correction based on the comparison; and

replacing a portion of the recognition result with the recognition correction.

16. A method of correcting incorrect text associated with recognition errors in computer-implemented speech recognition, the method comprising:

performing speech recognition on an utterance to produce a recognition result for the utterance;

receiving a selection of a word from the recognized utterance, the selection indicating a bound of a portion of the recognized utterance to be corrected; and

comparing an alternative transcript to the recognized utterance to be corrected, the comparing including;

selecting from the alternative transcript a test word that is not identical to the selected word and that occurs at a time that is nearest a time at which the selected word occurs, and

searching in time, through the recognized utterance relative to the selected word, and through the alternative transcript relative to the test word, until finding a word common to the recognized utterance and the alternative transcript.

17. The method of claim 16, further comprising:

producing a recognition correction based on the comparison;

replacing a portion of the recognition result with the recognition correction;

wherein the portion of the recognition result replaced includes at one bound a word indicated by the selection and extends for the duration of the recognition correction with which the portion is replaced.

18. A method of correcting incorrect text associated with recognition errors in computer-implemented speech recognition, the method comprising:

receiving a text document formed by recognizing speech utterances using a vocabulary;

receiving a general confusability matrix having one or more values each indicating a likelihood of confusion between a first phoneme and a second phoneme;

receiving corrected text that corresponds to misrecognized text from the text document;

generating a sequence of phonemes for the corrected text;

aligning the generated sequence of phonemes with phonemes of the misrecognized text;

adjusting one or more values of the general confusability matrix based on the alignment to form a specific confusability matrix; and

searching the text document for additional instances of the corrected text using the specific confusability matrix.

19. The method of claim 18 in which generating the sequence of phonemes for the corrected text includes determining whether the corrected text is in the vocabulary.

20. The method of claim 18 further comprising outputting the text document.

21. The method of claim 18 in which a list of recognition candidates is associated with each recognized speech utterance.

22. The method of claim 21 in which the empirical data comprises information relating to frequency characteristics of different phonemes.

23. The method of claim 21 in which the empirical data comprises information acquired during an adaptive training of a user.

24. The method of claim 18 in which generating the sequence of phonemes for the corrected text comprises using a phonetic alphabet.

25. The method of claim 18 further comprising generating the general confusability matrix using empirical data.

26. The method of claim 25 in which the empirical data comprises information relating to a rate of confusion of phonemes for a preselected population.

27. The method of claim 18 in which searching the text document for the corrected text comprises searching the text document for the sequence of phonemes for the corrected text.

28. The method of claim 18 in which searching the text document for the corrected text comprises searching the text document for a sequence of phonemes that is likely to be confused with the sequence of phonemes for the corrected text.

29. The method of claim 18 in which searching the text document for the corrected text comprises scoring a portion of the text document and comparing the score of the portion to an empirically determined threshold value to determine whether the portion of the text document includes a word that is not in the vocabulary.

30. The method of claim 29 further comprising outputting a result if it is determined that the portion of the text document includes a word that is not in the vocabulary.

31. The method of claim 30 in which outputting the result comprises highlighting the portion of the text document.

32. The method of claim 30 in which outputting the result comprises re-recognizing the portion of the text document.