Need math formulas for IPP GMM-functions. Like ippsLogGaussMixture_32f_D2() and others.

[pavelkolodin]

Trying to replace my customimplementation of calculation of the likelihood probability for the Gaussian mixtureto IPP procedures, like ippsLogGaussMixture_32f_D2, but i am not satisfied and not undestand results of this function, so i want to know how it treat my input data, so i need formulas.

[Jeffrey_M_Intel1]

The ippsLogGaussMixture functionsare part of the speech recognition domain. As you may have already seen, this domain wasremoved from IPP 7.0. For more information seethis thread.

We may be able to provide additional information on using these functions in terms of continuing <=6.1 support. Please let us know if you are still interested.

Best regards,

Jeff

[pavelkolodin]

Still interested :)

Thank you. Pavel.

[SergeyKostrov]

There is a description with formulas for the 'ippsLagGaussMixture...' functionsinan 'Intel IPP Signal Processing' Reference Manual volume 1 ( aspdf file ).
Did you have a chance to look at it?

[Jeffrey_M_Intel1]

These functions were implemented a long time ago. In case the manual is not sufficient we're looking foradditional reference material. We'll have more updates soon.

[SergeyKostrov]

Quoting Jeffrey Mcallister (Intel)

These functions were implemented a long time ago.

[SergeyK] That's right andit wasabout 10 years ago.

In case the manual is not sufficient we're looking foradditional reference material.We'll have more updates soon.

[SergeyK] Even if a theory ( for these functions )has not changed it is a good idea. Please keep everybody informed.

Best regards,
Sergey

[Jeffrey_M_Intel1]

For more information on the theory behind ippsLogGaussMixture, please see

Douglas Reynolds Gaussian Mixture Models MIT Lincoln Laboratory, 244 Wood St., Lexington, MA 02140, USA

There are several places to find this article, here is one:

http://www.ll.mit.edu/mission/communications/ist/publications/0802_Reynolds_Biometrics-GMM.pdf

You may also be interested in

SRIVASTAVA, S. Fast gaussian evaluations in large vocabulary continuous speech recognition. M.S. Thesis, Department of Electrical and Computer Engineering, Mississippi State University, Oct. 2002

In general, IPP's GMM implementation should be similar to others in terms of parameters and functionality.


Best regards,

Jeff

[SergeyKostrov]

Page 8-120 in IPP Signal Processing Manual:

[SergeyKostrov]

Page 8-121 in IPP Signal Processing Manual: