Automating tuning of speech recognition systems

• US 20030139925A1
• Filed: 12/31/2001
• Published: 07/24/2003
• Est. Priority Date: 12/31/2001
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Anderson, Andrew V., Bennett, Steven M.
• Granted Patent: US 7,203,644 B2
• Time in Patent Office: Days
• US Class Current: 704/231
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 2015/0631 Creating reference templates; Clustering
• CPC Class Codes: G10L 2015/225 Feedback of the input speech

Abstract

A speech recognition system. The system includes at least one recognizer to produce output signals from audio input signals and a feedback module to collect feedback data generated from conversion of the audio input signals to output signals. The system stores the feedback data and may also be adapted to use the feedback data as part of a training set for recognizers in the system, as part of a predictive mechanism, or both.

Claims

1. A speech recognition system, comprising:

at least one recognizer to produce output signals from audio input signals; and

a feedback module to generate feedback data.

2. The speech recognition system of claim 1, wherein the speech recognition system further comprises a controller operable to coordinate production of the output signals.

3. The speech recognition system of claim 2, wherein the controller is adaptable to provide the feedback data to the recognizer wherein the recognizer is operable to receive the feedback data.

4. The speech recognition system of claim 2, wherein the controller is adaptable to modify grammar files based on the feedback data.

5. The speech recognition system of claim 2, wherein the controller is adaptable to store the feedback data in a storage.

6. The speech recognition system of claim 2, wherein the controller is adaptable to update speech models based on the feedback data.

7. The speech recognition system of claim 1, wherein at least one recognizer further comprises multiple recognizers and a predictor to select a best performing recognizer from the multiple recognizers based upon the feedback data.

8. The speech recognition system of claim 1, wherein the output signals correspond to one of the group comprised of:

text, and command signals.

9. The system of claim 1, where the feedback module is adapted to generate feedback data based on internal analysis of at least one of the group comprised of:

grammar files, dialog progression, and output signals.

10. The system of claim 1, wherein the feedback module is adapted to generate feedback data based on external inputs comprised of at least one of the group comprised of:

annotated grammar files and information received through an application programming interface.

11. A speech recognition system, comprising:

at least one speech recognizer to convert audio input signals to output signals, wherein the speech recognizer is adapted to receive feedback data and adjust operation based upon the feedback data.

12. The speech recognition system of claim 11, wherein the system further comprises a controller operable to provide the feedback data to the recognizer.

13. The speech recognition system of claim 11, wherein the controller is adaptable to provide the feedback data to the recognizer.

14. The speech recognition system of claim 13, wherein the speech recognizer receives the feedback data in a manner of one of the group comprised of:

real-time, and off-line.

15. The speech recognition system of claim 11, wherein the speech recognition system further comprises a feedback module to collect feedback data.

16. A method of generating speech recognition feedback data, the method comprising:

converting an audio input signal to an output signal;

estimating a correctness measure wherein the correctness measure expresses if the output signal is a correct representation of the audio input signal; and

forming a feedback data element wherein the element consists of at least one of the audio input signal, the output signal, and the correctness measure.

17. The method of claim 16, wherein the method further comprises storing the feedback data element.

18. The method of claim 17, wherein storing the feedback data element further comprises storing one of the group comprised of:

only those feedback data elements for which the correction measure indicates that the output signal was not correct and those feedback data elements for which the correction measure indicates that the output signal was correct.

19. The method of claim 16, wherein the feedback data is filtered according to a criteria.

20. The method of claim 16, wherein the method further comprises utilizing the feedback data element, wherein utilizing comprises at least one of the group comprised of:

modifying a grammar file based on the feedback data, updating speech models based on the feedback data and updating a prediction mechanisms based on the feedback data.

21. The method of claim 16, wherein the method further comprises providing the feedback data element to a speech recognition system in which the feedback data is being collected.

22. The method of claim 16 wherein estimating a correctness measure further comprises at least one from a group comprised of:

receiving information through an application programming interface, analyzing grammar files, analyzing the output signal and analysis of the progression of the dialog.

23. The method of claim 16, wherein the method further comprises:

assigning an identifier to the audio input signal; and

including the identifier as part of the feedback data element.

24. The method of claim 16, wherein the method further comprises:

identifying relevant contextual information; and

including the relevant contextual information as part of the feedback data element.

25. An article including machine-readable code that, when executed, causes a machine to:

convert an audio input signal to an output signal;

estimate a correctness measure wherein the correctness measure expresses if the output signal is a correct representation of the audio input signal; and

form a feedback data element wherein the element consists of at least one of the audio input signal, the output signal, and the correctness measure.

26. The article of claim 25, wherein the article contains further machine-readable code that, when executed, causes the machine to provide the feedback data element to a speech recognition system in which feedback data is being collected.

27. The article of claim 25, wherein the code that, when executed, causes the machine to provide the feedback data element further causes the machine to utilize the feedback data element wherein utilizing the feedback data comprises at least one of the group comprising:

modifying a grammar file based on the feedback data, updating speech models based on the feedback data and updating a prediction mechanisms based on the feedback data.

28. The article of claim 25, wherein the article contains further machine-readable code that, when executed, causes the machine to store only those audio input signals for which the correction status indicates that a correction to the output signal was necessary.

29. The article of claim 25, wherein the article contains further machine-readable code that, when executed, causes the machine to store only those audio input signals for which the correction status indicates that no correction to the output signal was necessary.