Program for ML-Based Parameter Generation
* FOR LITTLE ENDIAN

usage: mlpg [options...] [inweightfile] [inmeanfile] [incovfile] [outfile]
options:
        -dim dim[28]                    : dimension of joint vector
        -clsnum clsnum[1]               : number of mixtures
        -maxitnum maxitnum[20]          : maximum number of iterations
        -thresh thresh[0.00001]         : threshold of increasing likelihood
        -vthresh vthresh[0.1]           : threshold of variance
        -dynwinf dynwinf                : window file for dynamic feature
        -stfile stfile                  : static sequence files
        -vmfile vmfile                  : mean of variance files
        -vvfile vvfile                  : variance of variance files
        -flkfile flkfile                : frame-based likelihood files
        -dia                            : using diagonal covariance
        -sm                             : smoothing by moving average
        -last                           : using variance after iteration
        -nmsg                           : no message
        -help                           : display this message

Three input files are binary files of weight sequences, mean vector
sequences and covariance matrices. The output file is a binary file of a
static feature sequence generated by a GMM-based PDF sequence. The number
of dimensions of the feature vector is changed by -dim.
* Note that the feature vector consists of static and dynamic
features.

A binary file of a delta window is necessary and it is used by
-dynwinf. The number of mixtures is changed by -clsnum. If -dia is
used, diagonal covariance matrices are used for the parameter generation.

A time sequence of the static feature vector are generated with EM
algorithm. Specific static vectors can be used as initial values for
EM.  A binary files of the static vectors is used by -stfile.  The
number of iterations is changed by -maxitnum, and a threshold of
likelihood for terminating the iteration is changed by -thresh.

A binary file of frame-based likelihoods is generated by -flkfile.

If -sm is used, the smoothing of generated vectors by a triangular
window with 3 point length is performed.

### Note that THE FOLLOWING PART IS STILL IN DEVELOPMENT. ###
ML-based parameter generation can be performed while considering
global variance of generated features. Binary files of Mean and
diagonal covariance files are necessary and they are used by -vmfile
and -vvfile. A steepest descent algorithm is performed. A threshold
for terminating the iteration is changed by -vthresh. If -last is
used, this parameter generation algorithm is applied to only the final
step of EM algorithm. This algorithm currently works only when using
diagonal covariance matrices.
##############################################################

Binary data has double-type format.

Some messages are printed while executing this program. If you
don't need those messages, use -nmsg.


<EXAMPLE>
% mlpg \
	-dim 48 \
        -clsnum 64 \
        -dynwinf input.win \
        -stfile input.vecs \
        -flkfile output.flk \
        -dia \
        -sm \
        -nmsg \
	input.wseq \
	input.mseq \
	input.covs \
	output.vecs


Tomoki Toda (tomoki@ics.nitech.ac.jp)


==================================================
The current copyright is

/*********************************************************************/
/*                                                                   */
/*            Nagoya Institute of Technology, Aichi, Japan,          */
/*                                and                                */
/*             Carnegie Mellon University, Pittsburgh, PA            */
/*                      Copyright (c) 2003-2004                      */
/*                        All Rights Reserved.                       */
/*                                                                   */
/*  Permission is hereby granted, free of charge, to use and         */
/*  distribute this software and its documentation without           */
/*  restriction, including without limitation the rights to use,     */
/*  copy, modify, merge, publish, distribute, sublicense, and/or     */
/*  sell copies of this work, and to permit persons to whom this     */
/*  work is furnished to do so, subject to the following conditions: */
/*                                                                   */
/*    1. The code must retain the above copyright notice, this list  */
/*       of conditions and the following disclaimer.                 */
/*    2. Any modifications must be clearly marked as such.           */
/*    3. Original authors' names are not deleted.                    */
/*                                                                   */    
/*  NAGOYA INSTITUTE OF TECHNOLOGY, CARNEGIE MELLON UNIVERSITY, AND  */
/*  THE CONTRIBUTORS TO THIS WORK DISCLAIM ALL WARRANTIES WITH       */
/*  REGARD TO THIS SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF     */
/*  MERCHANTABILITY AND FITNESS, IN NO EVENT SHALL NAGOYA INSTITUTE  */
/*  OF TECHNOLOGY, CARNEGIE MELLON UNIVERSITY, NOR THE CONTRIBUTORS  */
/*  BE LIABLE FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR  */
/*  ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR       */
/*  PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER   */
/*  TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE    */
/*  OR PERFORMANCE OF THIS SOFTWARE.                                 */
/*                                                                   */
/*********************************************************************/
/*                                                                   */
/*          Author :  Tomoki Toda (tomoki@ics.nitech.ac.jp)          */
/*          Date   :  June 2004                                      */
/*                                                                   */
/*-------------------------------------------------------------------*/
/*                                                                   */
/*  ML-Based Parameter Generation                                    */
/*                                                                   */
/*-------------------------------------------------------------------*/
