METHOD FOR ASSESSING PRONUNCIATION ABILITIES

• US 20090171661A1
• Filed: 06/27/2008
• Published: 07/02/2009
• Est. Priority Date: 12/28/2007
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Pant, Himanshu, Jayadeva, Joshi, Sachindra, Verma, Ashish
• Granted Patent: US 8,271,281 B2
• Time in Patent Office: Days
• US Class Current: 704/250
• CPC Class Codes: G09B 19/04 Speaking with audible presentation of the material to be studied G09B5/04
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence

Abstract

Techniques for assessing pronunciation abilities of a user are provided. The techniques include recording a sentence spoken by a user, performing a classification of the spoken sentence, wherein the classification is performed with respect to at least one N-ordered class, and wherein the spoken sentence is represented by a set of at least one acoustic feature extracted from the spoken sentence, and determining a score based on the classification, wherein the score is used to determine an optimal set of at least one question to assess pronunciation ability of the user without human intervention.

Claims

1. A method for assessing pronunciation abilities of a user, comprising the steps of:

recording a sentence spoken by a user;

performing a classification of the spoken sentence, wherein the classification is performed with respect to at least one N-ordered class, and wherein the spoken sentence is represented by a set of at least one acoustic feature extracted from the spoken sentence; and

determining a score based on the classification, wherein the score is used to determine an optimal set of at least one question to assess pronunciation ability of the user without human intervention.

2. The method of claim 1, wherein the at least one N-ordered class comprises a plurality of different classes.

3. The method of claim 1, wherein the score is used to determine a set of at least one sentence that is most discriminating for a given class pair when spoken by a user.

4. The method of claim 1, wherein the score is used to determine a minimal sequence of at least one sentence to be spoken by a user to classify the user in one of the at least one given N-ordered class in such a way that it matches the class assigned by a human expert.

5. The method of claim 1, wherein for each spoken sentence, a feature vector is generated and assessed on an integral scale of 1 to K by at least one human assessor.

6. The method of claim 1, further comprising performing a forced viterbi decoding using a speech recognition model, wherein the model is trained on at least one expert speaker using a training text that a candidate is given to read.

7. The method of claim 6, wherein the model is used for at least one of aligning text in terms of phoneme and time stamp, decoding a disfluency, and giving the confidence of each phoneme and each word.

8. The method of claim 1, further comprising discovering at least one of a substitution error, an insertion error and a deletion error.

9. The method of claim 1, further comprising using training data, wherein the training data comprises a group of at least one candidate assessed on a set of at least one sentence, and wherein each sentence is given a rating of 1 to N by at least one human expert.

10. The method of claim 9, further comprising using the training data to generate K−

1 classifiers corresponding to a binary search tree for K classes.

11. The method of claim 10, further comprising, for each of the K−

1 classifiers, determining which sentence is most discriminative.

12. A computer program product comprising a computer readable medium having computer readable program code for assessing pronunciation abilities of a user, said computer program product including:

computer readable program code for recording a sentence spoken by a user;

computer readable program code for performing a classification of the spoken sentence, wherein the classification is performed with respect to at least one N-ordered class, and wherein the spoken sentence is represented by a set of at least one acoustic feature extracted from the spoken sentence; and

computer readable program code for determining a score based on the classification, wherein the score is used to determine an optimal set of at least one question to assess pronunciation ability of the user without human intervention.

13. The computer program product of claim 12, wherein the computer readable program code for determining a score comprises computer readable program code for using the score to determine a set of at least one sentence that is most discriminating for a given class pair when spoken by a user.

14. The computer program product of claim 12, wherein the computer readable program code for determining a score comprises computer readable program code for using the score to determine a minimal sequence of at least one sentence to be spoken by a user to classify the user in one of the at least one given N-ordered class in such a way that it matches the class assigned by a human expert.

15. The computer program product of claim 12, further comprising computer readable program code for performing a forced viterbi decoding using a speech recognition model, wherein the model is trained on at least one expert speaker using a training text that a candidate is given to read.

16. The computer program product of claim 12, further comprising computer readable program code for using training data, wherein the training data comprises a group of at least one candidate assessed on a set of at least one sentence, and wherein each sentence is given a rating of 1 to N by at least one human expert.

17. A system for assessing pronunciation abilities of a user, comprising:

a memory; and

at least one processor coupled to said memory and operative to;

record a sentence spoken by a user;

perform a classification of the spoken sentence, wherein the classification is performed with respect to at least one N-ordered class, and wherein the spoken sentence is represented by a set of at least one acoustic feature extracted from the spoken sentence; and

determine a score based on the classification, wherein the score is used to determine an optimal set of at least one question to assess pronunciation ability of the user without human intervention.

18. The system of claim 17, wherein in determining a score, the at least one processor coupled to said memory is further operative to use the score to determine a set of at least one sentence that is most discriminating for a given class pair when spoken by a user.

19. The system of claim 17, wherein in determining a score, the at least one processor coupled to said memory is further operative to use the score to determine a minimal sequence of at least one sentence to be spoken by a user to classify the user in one of the at least one given N-ordered class in such a way that it matches the class assigned by a human expert.

20. The system of claim 17, wherein the at least one processor coupled to said memory is further operative to perform a forced viterbi decoding using a speech recognition model, wherein the model is trained on at least one expert speaker using a training text that a candidate is given to read.

21. The system of claim 17, wherein the at least one processor coupled to said memory is further operative to use training data, wherein the training data comprises a group of at least one candidate assessed on a set of at least one sentence, and wherein each sentence is given a rating of 1 to N by at least one human expert.