Method and system for matching speech data

• US 20070094020A1
• Filed: 10/20/2005
• Published: 04/26/2007
• Est. Priority Date: 10/20/2005
• Status: Active Grant

• Current Assignee: National Cheng KUNG University (Government of The Republic of China)
• Original Assignee: National Cheng KUNG University (Government of The Republic of China)
• Inventors: Lin, Po-Chuan, Wang, Jhing-Fa, Wen, Li-Chang
• Granted Patent: US 7,707,032 B2
• Time in Patent Office: Days
• US Class Current: 704/239
• CPC Class Codes: G10L 15/08 Speech classification or search

Abstract

A method and system used to determine the similarity between an input speech data and a sample speech data is provided. First, the input speech data is segmented into a plurality of input speech frames and the sample speech data is segmented into a plurality of sample speech frames. Then, the input speech frames and the sample speech frames are used to build a matching matrix, wherein the matching matrix comprises the distance values between each of the input speech frames and each of the sample speech frames. Next, the distance values are used to calculate a matching score. Finally, the similarity between the input speech data and the sample speech data is determined according to this matching score.

Claims

1. A method for matching speech data used to determine the similarity between an input speech data and a sample speech data, the method comprising:

segmenting the input speech data into a plurality of input speech frames;

segmenting the sample speech data into a plurality of sample speech frames;

building a matching matrix, wherein each element of the matching matrix corresponds to one of the input speech frames and one of the sample speech frames and indicates a distance value between the corresponding input speech frame and the corresponding sample speech frame;

calculating a matching score, wherein the matching score is obtained by calculating the distance value indicated in each element of the matching matrix; and

determining whether the input speech data and the sample speech data are similar according to the matching score.

2. The method as claimed in claim 1, wherein the distance value indicated in each element of the matching matrix is obtained by the dynamic timing warping (DTW) calculation.

3. The method as claimed in claim 1, wherein the process for calculating the matching score comprises:

calculating a row score, wherein the row score is obtained by calculating at least one distance value indicated in each row of elements of the matching matrix;

calculating a column score, wherein the column score is obtained by calculating at least one distance value indicated in each column of elements of the matching matrix; and

producing the matching score, wherein the matching score is obtained by calculating the row score and the column score.

4. The method as claimed in claim 3, wherein the row score is equal to the sum of a least distance value indicated in each row of elements of the matching matrix.

5. The method as claimed in claim 3, wherein the row score is equal to the sum of a least distance value indicated in each row of elements of the matching matrix plus the sum of a second least distance value indicated in each row of elements of the matching matrix.

6. The method as claimed in claim 3, wherein the row score is equal to a first product plus a second product, and the first product is equal to a first weighted value multiplied by the sum of a least distance value indicated in each row of elements of the matching matrix, and the second product is equal to a second weighted value multiplied by the sum of a second least distance value indicated in each row of elements of the matching matrix.

7. The method as claimed in claim 6, wherein the first weighted value is larger than the second weighted value.

8. The method as claimed in claim 3, wherein the column score is equal to the sum of a least distance value indicated in each column of elements of the matching matrix.

9. The method as claimed in claim 3, wherein the column score is equal to the sum of a least distance value indicated in each column of elements of the matching matrix plus the sum of a second least distance value indicated in each column of elements of the matching matrix.

10. The method as claimed in claim 3, wherein the column score is equal to a first product plus a second product, and the first product is equal to a third weighted value multiplied by the sum of a least distance value indicated in each column of elements of the matching matrix, and the second product is equal to a fourth weighted value multiplied by the sum of a second least distance value indicated in each column of elements of the matching matrix.

11. The method as claimed in claim 10, wherein the third weighted value is larger than the fourth weighted value.

12. The method as claimed in claim 3, wherein the matching score is equal to the sum of the row score and the column score.

13. The method as claimed in claim 3, wherein the matching score is equal to the product of the column score and a fifth weighted value, plus the product of the row score and a sixth weighted value.

14. The method as claimed in claim 3, wherein the matching score is equal to the product of the column score and a fifth weighted value plus the product of the row score, a sixth weighted value and a normalization parameter.

15. The method as claimed in claim 14, wherein the normalization parameter is equal to the amount of elements in each row of the matching matrix divided by the amount of elements in each column of the matching matrix.

16. The method as claimed in claim 3, wherein the matching score is equal to the product of the column score, a fifth weighted value and a normalization parameter, plus the product of the row score and a sixth weighted value.

17. The method as claimed in claim 16, wherein the normalization parameter is equal to the amount of elements in each column of the matching matrix divided by the amount of elements in each row of the matching matrix.

18. A speech data matching system used to determine the similarity between an input speech data and a sample speech data, the system comprising:

a speech data segmenting module used to segment the input speech data into a plurality of input speech frames and segment the sample speech data into a plurality of sample speech frames;

a matching matrix building module used to build a matching matrix, wherein each of the element of the matching matrix corresponds to one of the input speech frames and one of the sample speech frames and indicates a distance value between the corresponding input speech frame and the corresponding sample speech frame;

a matching score calculating module used to calculate a matching score, wherein the matching score is obtained by calculating the distance value indicated in each element of the matching matrix; and

a matching determining module used to determine whether the input speech data and the sample speech data are similar according to the matching score and then provide a result.

19. The speech data matching system as claimed in claim 18, wherein the matching matrix building module carries out the dynamic timing warping (DTW) calculation for obtaining the distance value indicated in each element of the matching matrix.

20. The speech data matching system as claimed in claim 18, wherein the matching score calculating module comprises:

a row score calculating module used to calculate a row score, wherein the row score calculating module calculates at least one distance value indicated in each row of elements of the matching matrix for obtaining the row score;

a column score calculating module used to calculate a column score, wherein the column score calculating module calculates at least one distance value indicated in each column of elements of the matching matrix for obtaining the column score; and

a matching score producing module used to produce the matching score, wherein the matching score producing module calculates the row score and the column score for obtaining the matching score.

21. The speech data matching system as claimed in claim 20, wherein the row score calculating module sums a least distance value indicated in each row of elements of the matching matrix for obtaining the row score.

22. The speech data matching system as claimed in claim 20, wherein the row score calculating module sums the sum of a least distance value indicated in each row of elements of the matching matrix and the sum of a second least distance value indicated in each row of elements of the matching matrix for obtaining the row score.

23. The speech data matching system as claimed in claim 20, wherein the row calculating module sums a first product and a second product for obtaining the row score, and the first product is equal to a first weighted value multiplied by the sum of a least distance value indicated in each row of elements of the matching matrix, and the second product is equal to a second weighted value multiplied by the sum of a second least distance value indicated in each row of elements of the matching matrix.

24. The speech data matching system as claimed in claim 23, wherein the first weighted value is larger than the second weighted value.

25. The speech data matching system as claimed in claim 20, wherein the column score calculating module sums up a least distance value indicated in each column of element of the matching matrix for obtaining the column score.

26. The speech data matching system as claimed in claim 20, wherein the column score calculating module sums up the sum of a least distance value indicated in each column of element of the matching matrix and the sum of a second least distance value indicated in each column of element of the matching matrix for obtaining the column score.

27. The speech data matching system as claimed in claim 20, wherein the column calculating module sums up a first product and a second product for obtaining the column score, and the first product is equal to a third weighted value multiplies the sum of a least distance value indicated in each column of element of the matching matrix and the second product is equal to a fourth weighted value multiplies the sum of a second least distance value indicated in each column of element of the matching matrix.

28. The speech data matching system as claimed in claim 27, wherein the third weighted value is larger than the fourth weighted value.

29. The speech data matching system as claimed in claim 20, wherein the matching score producing module sums the row score and the column score for obtaining the matching score.

30. The speech data matching system as claimed in claim 20, wherein the matching score producing module sums the product of the column score and a fifth weighted value and the product of the row score and a sixth weighted value for obtaining the matching score.

31. The speech data matching system as claimed in claim 20, wherein the matching score producing module sums the product of the column score and a fifth weighted value and the product of the row score, a sixth weighted value and a normalization parameter for obtaining the matching score.

32. The speech data matching system as claimed in claim 31, wherein the normalization parameter is equal to the amount of elements in each row of the matching matrix divided by the amount of elements in each column of the matching matrix.

33. The speech data matching system as claimed in claim 20, wherein the matching score producing module sums the product of the column score, a fifth weighted value and a normalization parameter and the product of the row score and a sixth weighted value for obtaining the matching score.

34. The speech data matching system as claimed in claim 33, wherein the normalization parameter is equal to the amount of elements in each column of the matching matrix divided by the amount of elements in each row of the matching matrix.