Error correction in speech recognition

• US 6,064,959 A
• Filed: 03/28/1997
• Issued: 05/16/2000
• Est. Priority Date: 03/28/1997
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: Dragon Systems, Inc. (Microsoft Corporation)
• Inventors: Roth, Robert, Van Even, Stijn, Gadbois, Gregory J., Young, Jonathan Hood, Dubach, Joev, Parmenter, David Wilsberg
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): Lerner, Martin
• Application Number:
  US08/825,534
• Time in Patent Office: 1,145 Days
• Field of Search: 704/10, 704/231, 704/243, 704/244, 704/251, 704/255, 704/256, 704/257, 704/275, 704/9
• US Class Current: 704/251
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 2015/228 of application context

Abstract

Incorrect text associated with recognition errors in computer-implemented speech recognition is corrected by performing speech recognition on an utterance to produce a recognition result for the utterance. When a correction command is identified in the recognition result for the utterance, corrected text is produced from a portion of the recognition result for the utterance.

Claims

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

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

identifying a correction command in the recognition result for the utterance; and

identifying corrected text from a portion of the recognition result for the utterance,

wherein the correction command indicates that the portion of the recognition result comprises a pronunciation of a word to be corrected.

2. The method of claim 1, further comprising replacing previously-generated incorrect text with the corrected text.

3. The method of claim 2, further comprising automatically selecting the previously-generated incorrect text to be replaced.

4. The method of claim 3 wherein the step of automatically selecting comprising re-recognizing previously-recognized speech corresponding to the previously-generated incorrect text using the corrected text.

5. The method of claim 1, wherein the step of identifying corrected text includes searching a dictionary using the portion of the recognition result.

6. The method of claim 1, wherein the step of identifying corrected text comprises identifying corrected text from a portion of the recognition result for the utterance and from a recognition result for a second utterance.

7. The method of claim 6, wherein the second utterance precedes the utterance.

8. The method of claim 6, wherein the second utterance follows the utterance.

9. The method of claim 1, wherein the step of identifying corrected text comprises using confused pronunciation matching, in which commonly-confused sounds are treated as a single sound, to identify text corresponding to the pronunciation.

10. The method of claim 9, wherein the confused pronunciation matching comprises using the pronunciation to search a confused pronunciation dictionary.

11. The method of claim 9, wherein the confused pronunciation matching comprises using the pronunciation to search a pronunciation dictionary for confused pronunciation matches.

12. The method of claim 9, wherein the confused pronunciation matching comprises using a phonetic tree to search a pronunciation dictionary.

13. The method of claim 9, further comprising generating a list of confused pronunciation matches and identifying the corrected text as a selection from the list of confused pronunciation matches.

14. The method of claim 13, further comprising using the list of confused pronunciation matches to re-recognize previously-recognized speech so as to determine the corrected text.

15. The method of claim 13, further comprising displaying text corresponding to the list of confused pronunciation matches to a user and obtaining the selection from the user.

16. The method of claim 1, the method further comprises:

using an active vocabulary when performing speech recognition,

using a backup dictionary when identifying the corrected text, and

if the active vocabulary does not contain the corrected text, adding the corrected text to the active vocabulary.

17. A method for recognizing a spelling of a word in computer-implemented speech recognition, comprising:

performing speech recognition on an utterance to produce recognition results;

identifying a spelling command in the recognition results, wherein the spelling command indicates that a portion of the utterance comprises a spelling;

producing the spelling by searching a dictionary using the recognition results,

wherein producing the spelling comprises using confused spelling matching, in which commonly-confused letters are treated as a single letter, to identify the spelling corresponding to the portion of the utterance.

18. The method of claim 17, wherein the dictionary is a confused spelling dictionary and the confused spelling matching comprises using the spelling to search the confused spelling dictionary.

19. The method of claim 17, wherein the dictionary is a spelling dictionary and the confused spelling matching comprises using the spelling to search the spelling dictionary for confused spelling matches.

20. The method of claim 17, further comprising generating a list of confused spelling matches and identifying the text corresponding to the spelling as a selection from the list of confused spelling matches.

21. The method of claim 20, further comprising displaying the list of confused spelling matches to a user and obtaining the selection from the user.

22. The method of claim 17, wherein confused spelling matching comprises using characteristics of a speaker'"'"'s pronunciation to search the dictionary for confused spelling matches.