Method and apparatus for performing observation probability calculations

• US 7,356,466 B2
• Filed: 06/20/2003
• Issued: 04/08/2008
• Est. Priority Date: 06/28/2002
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: Kim, Sung-Jae, Kim, Tae-Su, Park, Hyun-Woo, Min, Byung-Ho, Jang, Ho-Rang, Hong, Keun-Cheol
• Primary Examiner(s): OPSASNICK, MICHAEL N
• Application Number:
  US10/465,556
• Publication Number:
  US 20040002861A1
• Time in Patent Office: 1,754 Days
• Field of Search: 704/242, 704236-240
• US Class Current: 704/240
• CPC Class Codes: G10L 15/144 Training of HMMs

Abstract

A method and apparatus for calculating an observation probability includes a first operation unit that subtracts a mean of a first plurality of parameters of an input voice signal from a second parameter of an input voice signal, and multiplies the subtraction result to obtain a first output. The first output is squared and accumulated N times in a second operation unit to obtain a second output. A third operation unit subtracts a given weighted value from the second output to obtain a third output, and a comparator stores the third output for a comparator stores the third output in order to extract L outputs therefrom, and stores the L extracted outputs based on an order of magnitude of the extracted L outputs.

Claims

1. An observation probability calculation apparatus, comprising:

a first operation unit subtracting a mean of first parameters extracted from representative phonemes from one of a plurality of second parameters extracted from an input voice signal and multiplying the result of the subtraction by a distribution degree of the mean to obtain a first output;

a second operation unit squaring the first output of the first operation unit and accumulating the result of the squaring N times to obtain a second output;

a third operation unit subtracting a given weighted value from the second output of the second operation unit to obtain a third output; and

a comparator storing the third outputs of the third operation unit, extracting L outputs from the third outputs of the third operation unit, and storing the extracted L outputs in order of magnitude;

wherein the extracted L outputs are used for voice recognition.

2. The apparatus of claim 1, further comprising:

a first operation register storing the first output and outputting the stored first output to the second operation unit;

a second operation register storing the second output and outputting the stored second output to the third operation unit; and

a third operation register storing the third output and outputting the stored third output to the comparator.

3. The apparatus of claim 1, further comprising:

a plurality of registers storing the second parameter, mean, distribution degrees and weighted value.

4. The apparatus of claim 1, wherein the third output is a probabilistic distance between the mean of the plurality of first parameters and the one of the plurality of second parameters.

5. The apparatus of claim 1, wherein the second parameters include m bits, and

wherein the mean includes m bits.

6. The apparatus of claim 1, wherein N is 13.

7. A voice recognition device including the observation probability calculation apparatus of claim 1.

8. The apparatus of claim 1, wherein the comparator comprises:

first through L-th registers storing the extracted L outputs in order of magnitude; and

first through L-th index registers storing an order in which the extracted L outputs stored in the first through L-th registers are generated.

9. The apparatus of claim 8, wherein L is 3.

10. The apparatus of claim 1, wherein the first and second operation units perform an observation probability calculation (y) in a hidden Markov model algorithm according to:

$y=\underset{i=0}{\overset{1}{\sum <br><br>}}<br><br>\underset{j=0}{\overset{12}{\sum <br><br>}}<br><br>\mathrm{var}<br><br>\left[i\right]<br><br>\left[j\right]<br><br>{\left(\mathrm{Feature}<br><br>\left[i\right]<br><br>\left[j\right]-\mathrm{Mean}<br><br>\left[i\right]\left[j\right]\right)}^2$

wherein i is a factor representing the types of representative phonemes, j represents a number of the second parameters, and var denotes distribution degree of the mean.

11. The apparatus of claim 10, wherein var=1/(2 * σ

²).

12. A computer-implemented method of calculating an observation probability, comprising:

subtracting a mean of a plurality of first parameters extracted from a representative group of an input voice signal from one of a plurality of second parameters extracted from the input voice signal to obtain a result;

multiplying the result to obtain a first output;

squaring the first output; and

accumulating the squared first output according to a number of the representative group and a number of the second parameters to obtain an observation probability value;

wherein the observation probability value is used for voice recognition.

13. The method of claim 12, wherein the plurality of first parameters are extracted from representative phonemes of the input voice signal.

14. The method of claim 12, wherein the observation probability value is an accumulation of probabilities, and

wherein each probability is representative of a sequence of syllables.

15. An observation probability calculation apparatus calculating an observation probability according to the method of claim 12.

16. The method of claim 12, wherein multiplying includes multiplying the result by a distribution degree to obtain the first output.

17. The method of claim 16, further comprising:

subtracting a given weighted value from the observation probability value to obtain a third output;

wherein a value of the third output is inversely proportional to the observation probability value.

18. The method of claim 17, wherein the third output is a probabilistic difference between the mean of the plurality of first parameters and the one of the plurality of second parameters, and

wherein the observation probability value increases as the probabilistic difference decreases.

19. The method of claim 17, further comprising:

extracting L outputs from the third output; and

selecting an L output having a greatest order of magnitude as a revised observation probability value.

20. The method of claim 19, wherein the revised observation probability value is a greatest probability that a sequence of syllables is recognized as the input voice signal.

21. A computer-implemented method of recognizing a voice signal, comprising:

subtracting a mean of a plurality of first parameters extracted from a representative group of an input voice signal from one of a plurality of second parameters extracted from the input voice signal to obtain a result;

multiplying the result to obtain a first output;

squaring the first output;

accumulating the squared first output according to a number of the representative group and a number of the second parameters to obtain a second output;

subtracting a given weighted value from the second output to obtain a third output;

extracting L outputs from the third output; and

selecting an L output having a greatest order of magnitude, the selected L output corresponding to the recognized voice signal.

22. A voice recognition device recognizing a voice signal according to the method of claim 21.