Method and system for Gaussian probability data bit reduction and computation

• US 20070112566A1
• Filed: 11/12/2005
• Published: 05/17/2007
• Est. Priority Date: 11/12/2005
• Status: Active Grant

• Current Assignee: Sony Interactive Entertainment Inc. (Sony Group Corp.)
• Original Assignee: Sony Computer Entertainment Incorporated (Sony Group Corp.)
• Inventors: Chen, Ruxin
• Granted Patent: US 7,970,613 B2
• Time in Patent Office: Days
• US Class Current: 704/240
• CPC Class Codes: G10L 15/14 using statistical models, e.g. Hidden Markov Models [HMMs] G10L15/18 takes precedence
• CPC Class Codes: G10L 15/285 Memory allocation or algorithm optimisation to reduce hardware requirements

Abstract

Use of runtime memory may be reduced in a data processing algorithm that uses one or more probability distribution functions. Each probability distribution function may be characterized by one or more uncompressed mean values and one or more variance values. The uncompressed mean and variance values may be represented by α-bit floating point numbers, where α is an integer greater than 1. The probability distribution functions are converted to compressed probability functions having compressed mean and/or variance values represented as β-bit integers, where β is less than α, whereby the compressed mean and/or variance values occupy less memory space than the uncompressed mean and/or variance values. Portions of the data processing algorithm can be performed with the compressed mean and variance values.

Claims

1. A method for reducing use of runtime memory in a data processing algorithm that uses one or more probability distribution functions, where each probability distribution function is characterized by one or more uncompressed mean values and one or more variance values, wherein the uncompressed mean and variance values are represented by α

-bit floating point numbers, where a is an integer greater than 1, the method comprising the steps of;

converting the probability distribution functions to compressed probability functions having compressed mean and/or variance values represented as β

-bit integers, where β

is less than α

, whereby the compressed mean and/or variance values occupy less memory space than the uncompressed mean and/or variance values; and

performing one or more portions of the data processing algorithm with the compressed mean and variance values.

2. The method of claim 1 wherein converting the probability distribution functions to compressed probability functions includes compressing values for one or more observable features from α

-bit floating point type to β

-bit integer type.

3. The method of claim 1 wherein α

is 32 and β

is 8.

4. The method of claim 1 wherein the one or more probability distribution functions include one or more Gaussian probability distribution functions.

5. The method of claim 4 wherein the Gaussian probability distribution functions include M Gaussian functions of N different observable features x₀ . . . x_n, where n=N−

1, each Gaussian function representing the probability distribution for a different one of M possible Gaussians, each Gaussian function having the form;

6. The method of claim 5 wherein the compressed mean values have the form:

{circumflex over (μ

)}_ki=φ

((μ

_ki-b_i)•

c_i),

where b_i is a centroid of the means for a given observable feature for all possible Gaussians,

where c_i is a variance for a given observable feature for all possible Gaussians,

where, for x=(μ

_ki-b_i)•

c_i, and Θ

=2^β

−

1,

$\mathrm{\varphi <br><br>}<br><br>\left(x\right)=\{\begin{array}{c}\mathrm{\Theta <br><br>\_if}<br><br>\mathrm{\_x}><br><br>\mathrm{\Theta <br><br>}\\ -\mathrm{\Theta <br><br>\_if}<br><br>\mathrm{\_x}<<br><br>-\mathrm{\Theta <br><br>}\end{array}$

and φ

(x)=a fixed point representation of x otherwise.

7. The method of claim 6 wherein

8. The method of claim 6 wherein

9. The method of claim 6 wherein converting the probability distribution functions to compressed probability functions includes compressing values for one or more observable features x₀ . . . x_n from α

-bit floating point type to compressed observable feature values of β

-bit integer type.

10. The method of claim 9 wherein the compressed observable feature values are given by {circumflex over (x)}_i=φ

((x_i−

b_i)•

c_i),where, for x=(x_ki−

b_i)•

c_i, and Θ

=2⁶² −

1,

11. The method of claim 10 wherein

12. The method of claim 10 wherein

13. The method of claim 5 wherein the compressed variances have the form

14. The method of claim 13 wherein λ

is equal to about 30.

15. The method of claim 13 wherein

16. The method of claim 5 wherein the compressed probability functions have the form

17. The method of claim 16, further comprising, empirically determining a value of λ

that optimizes recognition accuracy.

18. The method of claim 16 wherein A is an integer between 1 and 100.

19. The method of claim 16 wherein α

=32 β

=8 and λ

=30.

20. The method of claim 1 wherein the data processing algorithm is a recognition algorithm.

21. The method of claim 20 wherein the recognition algorithm includes observing N features of a signal, where N is an integer greater than 1, computing a probability for each of M possible Gaussians, where M is an integer greater than 1, and selecting a maximum probability from among the probabilities for the M possible Gaussians.

22. The method of claim 20 wherein observing the N features includes converting a time domain signal to a cepstrum characterized by a plurality of cepstral coefficients wherein the N features include selected cepstral coefficients.

23. The method of claim 20 wherein the recognition algorithm is a speech recognition algorithm or optical character recognition algorithm.

24. The method of claim 20 wherein the recognition algorithm uses a hidden Markov model.

25. A signal recognition apparatus, comprising:

a signal processor configured to observe N different features of a signal and set up M different probability functions of the N different features, each probability function representing a probability of a different one of M possible Gaussians of a portion of the signal, where each probability distribution function is characterized by one or more uncompressed mean values and one or more variance values, wherein the uncompressed mean and variance values are represented by α

-bit floating point numbers, where a is an integer greater than 1,

wherein the signal processor is configured to convert the probability distribution functions to compressed probability functions having compressed mean and/or variance values represented as β

-bit integers, where β

is less than α

, whereby the compressed mean and/or variance values occupy less memory than the uncompressed mean and/or variance values,

wherein the signal processor is configured to calculate a probability for each of the M possible Gaussians using the compressed probability functions, and

wherein the signal processor is configured to determine a most probable Gaussian from the calculated probabilities for the M possible Gaussians.

26. The apparatus of claim 25 wherein the one or more probability distribution functions include one or more Gaussian probability distribution functions.

27. The apparatus of claim 26 wherein the Gaussian probability distribution functions include M Gaussian functions of N different observable, features x₀ / / / x_n, where n=N−

1, each Gaussian function representing the probability distribution for a different one of M possible Gaussians, each Gaussian function having the form;

28. The apparatus of claim 27 wherein the compressed mean values have the form:

{circumflex over (μ

)}_ki=φ

((μ

_ki−

b_i)•

c_i),

where b_i is a centroid of the means for a given observable feature for all possible Gaussians,

where c_i is a variance for a given observable feature for all possible Gaussians,

where, for x=(μ

_ki−

b_i)•

c_i, and Θ

=2^β

−

1,

$\mathrm{\varphi <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>\left(x\right)=\{\begin{array}{c}\mathrm{\Theta <br><br>\_if}<br><br>\mathrm{\_x}><br><br>\mathrm{\Theta <br><br>}\\ -\mathrm{\Theta <br><br>\_if}<br><br>\mathrm{\_x}<<br><br>-\mathrm{\Theta <br><br>}\end{array}$

and φ

(x)=a fixed point representation of x otherwise.

29. The apparatus of claim 28 wherein

30. The apparatus of claim 28 wherein

31. The apparatus of claim 28 wherein converting the probability distribution functions to compressed probability functions includes compressing values for one or more observable features x₀. . . x_n from α

-bit floating point type to compressed observable feature values of β

-bit integer type.

32. The apparatus of claim 31 wherein the compressed observable feature values are given by {circumflex over (x)}_i=φ

((x_i−

b_i)•

c_i), where, for x=(x_ki−

b_i)•

c_i, and Θ

=2^β

−

1,

33. The apparatus of claim 32 wherein

34. The apparatus of claim 32 wherein

35. The apparatus of claim 27 wherein the compressed variances have the form

36. The apparatus of claim 35 wherein λ

is equal to about 30.

37. The apparatus of claim 35 wherein

38. The apparatus of claim 27 wherein the compressed probability functions have the form

39. The apparatus of claim 38, further comprising, a value of λ

has been empirically determined to optimize recognition accuracy.

40. An apparatus for reducing use of runtime memory in a data processing algorithm that uses one or more probability distribution functions, where each probability distribution function is characterized by one or more uncompressed mean values and one or more variance values, wherein the uncompressed mean and variance values are represented by α

-bit floating point numbers, where a is an integer greater than 1, the apparatus comprising;

means for converting the probability distribution functions to compressed probability functions having compressed mean and/or variance values represented as β

-bit integers, where β

is less than α

, whereby the compressed mean and/or variance values occupy less memory than the uncompressed mean and/or variance values; and

means for performing one or more portions of the data processing algorithm with the compressed mean and variance values.