User adaptive speech recognition method and apparatus

• US 20060200347A1
• Filed: 02/16/2006
• Published: 09/07/2006
• Est. Priority Date: 03/07/2005
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: Kim, Jeong-su, Kim, Jung-eun
• Granted Patent: US 7,996,218 B2
• Time in Patent Office: Days
• US Class Current: 704/236
• CPC Class Codes: G10L 15/08 Speech classification or search
• 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

A user adaptive speech recognition method and apparatus is disclosed that controls user confirmation of a recognition candidate using a new threshold value adapted to a user. The user adaptive speech recognition method includes calculating a confidence score of a recognition candidate according to the result of speech recognition, setting a new threshold value adapted to the user based on a result of user confirmation of the recognition candidate and the confidence score of the recognition candidate, and outputting a corresponding recognition candidate as a result of the speech recognition if the calculated confidence score is higher than the new threshold value. Thus, the need for user confirmation of the result of speech recognition is reduced and the probability of speech recognition success is increased.

Claims

1. A user adaptive speech recognition method comprising:

calculating a confidence score of a recognition candidate according to a result of speech recognition of speech input from a user;

setting a new threshold value adapted to the user based on a result of user confirmation of the recognition candidate and the confidence score of the recognition candidate; and

outputting a corresponding recognition candidate as the result of the speech recognition when the calculated confidence score is higher than the new threshold value.

2. The method as claimed in claim 1, wherein the setting a new threshold value comprises:

classifying recognition candidates selected as correct answers by the user into a first group and classifying recognition candidates selected as incorrect answers by the user into a second group; and

calculating the new threshold value having a value in a range that is higher than the highest confidence score of the recognition candidates classified into the second group and that is lower than the lowest confidence score of the recognition candidates classified into the first group, when a first confidence score section being the confidence scores of the recognition candidates classified into the first group and a second confidence score section being the confidence scores of the recognition candidates classified into the second group do not overlap.

3. The method as claimed in claim 2, wherein the new threshold value is an average value of the lowest confidence score of the recognition candidates classified into the first group and the highest confidence score of the recognition candidates classified into the second group.

4. The method as claimed in claim 2, further comprising calculating the new threshold value so as to have a value in a range that is higher than the lowest confidence score of the recognition candidates classified into the first group and lower than the highest confidence score of the recognition candidates classified into the second group, when the first confidence score section and the second confidence score section overlap.

5. The method as claimed in claim 4, wherein the new threshold value is included in a specified confidence score range, and

wherein the confidence score range is set so that a ratio of a number of the recognition candidates having confidence scores below a lower threshold value of the confidence score range among the recognition candidates classified into the first group to a number of recognition candidates having the confidence scores above an upper threshold value of the confidence score range among the recognition candidates classified into the second group is the same as a threshold ratio.

6. The method as claimed in claim 5, wherein the new threshold value is an average value of the lowest confidence score of a recognition candidate that is higher than the upper threshold value of the confidence score range and the highest confidence score of a recognition candidate that is lower than the lower threshold value of the confidence score range.

7. The method as claimed in claim 1, wherein the setting a new threshold value is performed when the number of speech inputs is greater than a threshold number of speech inputs.

14. The apparatus as claimed in claim 1, wherein the threshold value setup unit sets the new threshold value when the number of speech inputs is greater than a threshold number of speech inputs.

8. A user adaptive speech recognition apparatus comprising:

a confidence score calculation unit calculating a confidence score of a recognition candidate according to the result of speech recognition of speech input by a user;

a threshold value setup unit setting a new threshold value adapted to the user based on a result of user confirmation of the recognition candidate and the confidence score of the recognition candidate; and

a control unit outputting a corresponding recognition candidate as the result of the speech recognition when the calculated confidence score is higher than the new threshold value.

9. The apparatus as claimed in claim 8, wherein the threshold value setup unit comprises:

a classification unit classifying recognition candidates selected as correct answers by the user into a first group and classifying recognition candidates selected as incorrect answers by the user into a second group; and

a threshold value calculation unit calculating the new threshold value having a value in a range that is higher than the highest confidence score of the recognition candidates classified into the second group and that is lower than the lowest confidence score of the recognition candidates classified into the first group, when a first confidence score section being the confidence scores of the recognition candidates classified into the first group and a second confidence score section being the confidence scores of the recognition candidates classified into the second group do not overlap.

10. The apparatus as claimed in claim 9, wherein the new threshold value is an average value of the lowest confidence score of the recognition candidates classified into the first group and the highest confidence score of the recognition candidates classified into the second group.

11. The apparatus as claimed in claim 9, wherein the threshold value calculation unit calculates the new threshold value so as to have a value in a range that is higher than the lowest confidence score of the recognition candidates classified into the first group and that is lower than the highest confidence score of the recognition candidates classified into the second group, when the first confidence score section and the second confidence score section overlap.

12. The apparatus as claimed in claim 11, wherein the new threshold value is included in a specified confidence score range, and

wherein the confidence score range is set so that a ratio of a number of the recognition candidates having confidence scores below a lower threshold value of the confidence score range among the recognition candidates classified into the first group to a number of recognition candidates having the confidence scores above an upper threshold value of a confidence score range among the recognition candidates classified into the second group is the same as a threshold ratio.

13. The apparatus as claimed in claim 12, wherein the new threshold value is an average value of the lowest confidence score of a recognition candidate that is higher than the upper threshold value of the confidence score range and the highest confidence score of a recognition candidate that is lower than the lower threshold value of the confidence score range.

15. A computer-readable recording medium encoded with processing instructions for causing a processor to execute a user adaptive speech recognition method, the method comprising:

calculating a confidence score of a recognition candidate according to a result of speech recognition of speech input from a user;

setting a new threshold value adapted to the user based on a result of user confirmation of the recognition candidate and the confidence score of the recognition candidate; and

outputting a corresponding recognition candidate as the result of the speech recognition when the calculated confidence score is higher than the new threshold value.

16. A method of setting a new threshold value adapted to a user of a speech recognition apparatus, the method comprising:

determining whether a recognition candidate presented to the user is indicated by the user as correct;

classifying the recognition candidate based on a result of the determining, the recognition candidate being classified in a first group when the recognition candidate is correct, the recognition candidate being a classified in a second group when the recognition candidate is incorrect;

determining whether a range of probabilities of validity of the recognition candidates of the first group and a range of probabilities of validity of the recognition candidates of the second group overlap;

calculating the new threshold value in a range between the range of the probabilities of validity of the recognition candidates of the first group and the range of the probabilities of validity of the recognition candidates of the second group, when the range of the first group and the range of the second group do not overlap; and

calculating the new threshold value in a range where the range of the probabilities of validity of the recognition candidates of the first group and the range of the probabilities of validity of the recognition candidates of the second groups overlap, when the range of the first group and the range of the second group overlap.