Selecting one of multiple speech recognizers in a system based on performance predections resulting from experience

• US 6,996,525 B2
• Filed: 06/15/2001
• Issued: 02/07/2006
• Est. Priority Date: 06/15/2001
• Status: Expired due to Fees

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Anderson, Andrew V., Bennett, Steven M.
• Primary Examiner(s): Chawan, Vijay B.
• Application Number:
  US09/882,563
• Publication Number:
  US 20020194000A1
• Time in Patent Office: 1,698 Days
• Field of Search: 704/231, 704/233, 704/235, 704/240, 704/232, 704/251, 704/253, 704/256, 704/270.1, 704/270, 704/255, 704/704, 382/138
• US Class Current: 704/231
• CPC Class Codes: G10L 15/32 Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems

Abstract

A method for selecting a speech recognizer from a number of speech recognizers in a speech recognition system. The speech recognition system receives an audio stream from an application and derives enabling information. The speech recognition system then enables at least some of the speech recognizers and receives their results. It derives selection information and uses it to select the best speech recognizer and its results and returns those results back to the application.

Claims

1. A method for selecting a recognizer from a number of recognizers, the method comprising:

a) receiving an input stream;

b) deriving selection information, wherein the selection information includes performance-related information;

c) deriving enabling information, and using the enabling information to enable at least one selected recognizer to process the input stream;

d) using the selection information to select results from at least one enabled recognizer, and

e) returning the results to an application.

2. The method of claim 1, wherein the enabling information comprises at least one type of information from the group comprised of:

channel characteristics, device characteristics, user information, contextual information, dialog state, recognizer costs and performance history.

3. The method of claim 1, wherein the enabling information comprises recognizer costs.

4. The method as recited in claim 1, wherein the method further comprises receiving feedback and including the feedback in the selection information.

5. The method as recited in claim 1, wherein the derived enabling information including performance-related information comprises performance tracking using enabling information to select at least one recognizer.

6. The method as recited in claim 1, wherein the selecting results front one of the at least on enabled recognizers is based on selection information comprising performance predictors.

7. The method as recited in claim 6, wherein the performance predictors are selected from a group or predictors comprising channel characteristics, device characteristics, user information, contextual information, dialog state;

and individual-results confidence.

8. A method for selecting a recognizer from a number of recognizers, the method comprising

a) receiving an input stream;

b) deriving selection information, wherein the selection information includes performance-related information;

c) deriving enabling information, and using the enabling information to enable at least one selected recognizer to process the input stream;

d) using the selection information to select results from at least one enabled recognizer,

wherein the enabling information is used to enable a recognizer based upon its expected future performance; and

e) returning the results to in application.

9. An article containing machine-readable code that, when executed, causes the machine to

a) receive an input stream;

b) derive enabling information, wherein the enabling information includes performance-related information;

c) use the enabling information to select an enabled recognizer; and

d) return results from the enabled recognizer to an application.

10. The article of claim 9, the code causing the machine to derive enabling information includes code, that when executed, causes the machine to analyze the input stream for channel characteristics.

11. The article of claim 9, the code causing the machine to derive enabling information includes code, that when executed, causes the machine to receive contextual information associated with the input stream.

12. The method of claim 11, wherein contextual information comprises information from at least one item of information derived from the set of information comprising information related to the environment around the input stream, characteristics of a user generating die input stream, information derived from a call using network services, gender, age, ethnicity, information relating to the user'"'"'s first (native) language, personal information about the user, channel characteristics and device characteristics.

13. The method as recited in claim 12, wherein the contextual information is obtained dynamically.

14. The method as recited in claim 12, wherein the contextual information is predetermined.

15. The article of claim 9, the code including code, that when executed, causes the machine to receive feedback and include the feedback in the selection information.

16. The article as recited in claim 9, wherein the machine-readable code further causes the machine to receive feedback and includes the feedback in selection information.

17. The article as recited in claim 9, wherein the derived enabling information includes performance-related information comprising performance tracking, wherein the performance-tracking uses enabling formation to select at least one recognizer.

18. The article as recited in claim 9, wherein selecting results from one of the at least on enabled recognizers is based on selection information comprising performance predictors.