Enhanced stability prediction for incrementally generated speech recognition hypotheses

• US 20130110492A1
• Filed: 05/01/2012
• Published: 05/02/2013
• Est. Priority Date: 11/01/2011
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Gruenstein, Alexander H., McGraw, Ian C.
• Granted Patent: US 8,909,512 B2
• Time in Patent Office: Days
• US Class Current: 704/2
• CPC Class Codes: G03F 7/707 Chucks, e.g. chucking or un-chucking operations or structural details
• CPC Class Codes: G03F 7/70708 being electrostatic; Electrostatically deformable vacuum chucks
• CPC Class Codes: G03F 7/70875 Temperature, e.g. temperature control of masks or workpieces via control of stage temperature
• CPC Class Codes: G03F 7/70908 Hygiene, e.g. preventing apparatus pollution, mitigating effect of pollution or removing pollutants from apparatus
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 2015/223 Execution procedure of a spoken command
• CPC Class Codes: H01L 21/6831 using electrostatic chucks

Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for predicting the stability of speech recognition results. In one aspect, a method includes determining a length of time, or a number of occasions, in which a word has remained in an incremental speech recognizer'"'"'s top hypothesis, and assigning a stability metric to the word based on the length of time or number of occasions.

Claims

1. A computer-implemented method comprising:

receiving multiple, partial incremental speech recognition hypotheses that are each output by an incremental speech recognizer as a top partial incremental speech recognition hypothesis at a different point in time;

identifying a segment that occurs in a particular one of the multiple, partial incremental speech recognition hypotheses;

determining a quantity of consecutive, partial incremental speech recognition hypotheses that (i) are output by the incremental speech recognizer as the top partial incremental speech recognition hypotheses at different points in time immediately after the one particular partial incremental speech recognition hypothesis is output, and (ii) include the segment; and

assigning, by the one or more computers, a stability metric to the segment based on the quantity of consecutive, partial incremental speech recognition hypotheses that (i) are output by the incremental speech recognizer as the top partial incremental speech recognition hypotheses at different points in time immediately after the one particular partial incremental speech recognition hypothesis is output, and (ii) include the segment.

2. The method of claim 1, wherein assigning the stability metric to the segment is further based on a right context of the segment.

3. The method of claim 1, wherein the segment comprises a word or sub-word.

5. The method of claim 1, wherein the stability metric is assigned to the segment after the one particular partial incremental speech recognition hypothesis that includes the segment has been output by the incremental speech recognizer.

6. The method of claim 1, comprising:

receiving an audio signal corresponding to an utterance; and

performing incremental speech recognition on the audio signal to generate the multiple, partial incremental speech recognition hypotheses.

7. The method of claim 1, comprising:

displaying the segment that occurs in a particular one of the multiple, partial incremental speech recognition hypotheses on a user interface, the displayed segment having a visual characteristic;

determining that the stability metric of the segment satisfies a threshold; and

altering the visual characteristic of the displayed segment on the user interface only after determining that the stability metric satisfies the threshold.

8. The method of claim 1, comprising:

determining that the stability metric satisfies a threshold; and

translating the segment to a different language only after determining that the stability metric satisfies the threshold.

9. The method of claim 1, comprising:

determining that the stability metric satisfies a threshold; and

submitting the segment to a search engine as part of a search query only after determining that the stability metric satisfies the threshold.

10. The method of claim 1, comprising:

determining that the stability metric satisfies a threshold; and

displaying a representation of the segment on a user interface only after determining that the stability metric satisfies the threshold.

11. The method of claim 1, wherein the receiving, identifying, determining and assigning are performed by one or more computers.

31. The method of claim 1, wherein identifying the segment comprises identifying a particular segment that does not occur in a partial incremental speech recognition hypothesis that immediately precedes the one particular partial incremental speech recognition hypothesis.

32. The method of claim 1, comprising:

determining that the stability metric of the segment from the particular one of the multiple, partial incremental speech recognition hypotheses satisfies a threshold;

upon determining that the segment does not satisfy the threshold, displaying the segment on a user interface, the displayed segment having a first visual characteristic; and

upon determining that the segment satisfies the threshold, displaying the segment on the user interface, the displayed segment having a second visual characteristic different from the first visual characteristic.

4. (canceled)

12. A system comprising:

one or more computers and one or more storage devices storing instructions that are operable, if executed by the one or more computers, to cause the one or more computers to perform operations comprising;

receiving multiple, partial incremental speech recognition hypotheses that are each output by an incremental speech recognizer as a top partial incremental speech recognition hypothesis at a different point in time;

identifying a segment that occurs in a particular one of the multiple, partial incremental speech recognition hypotheses;

determining a quantity of consecutive, partial incremental speech recognition hypotheses that (i) are output by the incremental speech recognizer as the top partial incremental speech recognition hypotheses at different points in time immediately after the one particular partial incremental speech recognition hypothesis is output, and (ii) include the segment; and

assigning, by the one or more computers, a stability metric to the segment based on the quantity of consecutive, partial incremental speech recognition hypotheses that (i) are output by the incremental speech recognizer as the top partial incremental speech recognition hypotheses at different points in time immediately after the one particular partial incremental speech recognition hypothesis is output, and (ii) include the segment.

13. The system of claim 12, wherein assigning the stability metric to the segment is further based on a right context of the segment.

14. The system of claim 12, wherein the segment comprises a word or sub-word.

16. The system of claim 12, wherein the stability metric is assigned to the segment after the one particular partial incremental speech recognition hypothesis that includes the segment has been output by the incremental speech recognizer.

17. The system of claim 12, wherein the operations comprise:

receiving an audio signal corresponding to an utterance; and

performing incremental speech recognition on the audio signal to generate the multiple, partial incremental speech recognition hypotheses.

18. The system of claim 12, wherein the operations comprise:

displaying the segment that occurs in a particular one of the multiple, partial incremental speech recognition hypotheses on a user interface, the displayed segment having a visual characteristic;

determining that the stability metric of the segment satisfies a threshold; and

altering the visual characteristic of the displayed segment on the user interface only after determining that the stability metric satisfies the threshold.

20. The system of claim 12, wherein the operations comprise:

determining that the stability metric satisfies a threshold; and

submitting the segment to a search engine as part of a search query only after determining that the stability metric satisfies the threshold.

21. The method of claim 12, wherein the operations comprise:

determining that the stability metric satisfies a threshold; and

displaying a representation of the segment on a user interface only after determining that the stability metric satisfies the threshold.

33. The system of claim 12, wherein identifying the segment comprises identifying a particular segment that does not occur in a partial incremental speech recognition hypothesis that immediately precedes the one particular partial incremental speech recognition hypothesis.

15. (canceled)

19. (canceled)

22. A computer-readable storage device storing software comprising instructions executable by one or more computers, which, upon such execution, cause the one or more computers to perform operations comprising:

receiving multiple, partial incremental speech recognition hypotheses that are each output by an incremental speech recognizer as a top partial incremental speech recognition hypothesis at a different point in time;

identifying a segment that occurs in a particular one of the multiple, partial incremental speech recognition hypotheses;

determining a quantity of consecutive, partial incremental speech recognition hypotheses that (i) are output by the incremental speech recognizer as the top partial incremental speech recognition hypotheses at different points in time immediately after the one particular partial incremental speech recognition hypothesis is output, and (ii) include the segment; and

assigning, by the one or more computers, a stability metric to the segment based on the quantity of consecutive, partial incremental speech recognition hypotheses that (i) are output by the incremental speech recognizer as the top partial incremental speech recognition hypotheses at different points in time immediately after the one particular partial incremental speech recognition hypothesis is output, and (ii) include the segment.

23. The device of claim 22, wherein assigning the stability metric to the segment is further based on a right context of the segment.

24. The device of claim 22, wherein the segment comprises a word or sub-word.

26. The device of claim 22, where the stability metric is assigned to the segment after the one particular partial incremental speech recognition hypothesis that includes the segment has been output by the incremental speech recognizer.

27. The device of claim 22, wherein the operations comprise:

receiving an audio signal corresponding to an utterance; and

performing incremental speech recognition on the audio signal to generate the multiple, partial incremental speech recognition hypotheses.

28. The device of claim 22, wherein the operations comprise:

displaying the segment that occurs in a particular one of the multiple, partial incremental speech recognition hypotheses on a user interface, the displayed segment having a visual characteristic;

determining that the stability metric of the segment satisfies a threshold; and

altering the visual characteristic of the displayed segment on the user interface only after determining that the stability metric satisfies the threshold.

29. The device of claim 22, wherein the operations comprise:

determining that the stability metric satisfies a threshold; and

submitting the segment to a search engine as part of a search query only after determining that the stability metric satisfies the threshold.

34. The device of claim 22, wherein identifying the segment comprises identifying a particular segment that does not occur in a partial incremental speech recognition hypothesis that immediately precedes the one particular partial incremental speech recognition hypothesis.

25. (canceled)

30. (canceled)