[hts-users:02636] Re: GV param generation

["Heiga ZEN (Byung Ha CHUN)" <heiga.zen@xxxxxxxxxxxxxxxxx>]

Hi,

ÄÇÐËÓî wrote (2010/11/23 4:44):

> In the calculation of gradient using Newton-Raphson method, the code below:
> h = -w1 * w * R[i][1] -w2 * 2.0 * (.......);
> should it be:
> h = -w1 * w * R[i][1] -w2 * w * 2.0 * (.......);
> or not?

No, h = -w1 * w * R[i][1] -w2 * 2.0 * (.......); is correct.

w is to normalize the difference between dimensionalities of HMM and GV 
Gaussians.  It is given as

   const double w = 1.0/(pst->win.num*pst->T);

GV Gaussian is 1 dimensional.  HMM Gaussian is 3T (# of window is 3 and # of 
frames is T).  So w = 1/3T.  This normalization constant should be multiplied to 
HMM output prob related part of h.

Regards,

Heiga ZEN (Byung Ha CHUN)

======================================================================
NOTE: The information in this email and any attachments may be confidential
and/or legally privileged. This message may be read, copied and used only by
the intended recipient. If you are not the intended recipient, please
destroy this message, delete any copies held on your system and notify the
sender immediately.

Toshiba Research Europe Limited, registered in England and Wales (2519556).
Registered Office: 208 Cambridge Science Park, Milton Road, Cambridge
CB4 0GZ, England. Web: http://www.toshiba-europe.com/research/crl
======================================================================