SphinxTrain 1.0 Release-Candidate I
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

(Draft)

From initial check-in to SphinxTrain 1.0 RCI (Corresponds to Rev. 5826)

The current revision is based on SVN. In the CVS time, development of
SphinxTrain was very disorganized.  The list of changes was later
added by Arthur Chan.  The original intention might be distorted but
we thought this is better than changing the admin messages after the
fact.

Summary of this release:

From 2000, when SphinxTrain is first opened to the public, it was
regarded as merely an auxillary tool for the decoders (such as Sphinx
2, Sphinx3 and later Sphinx 4) .  However, many users in the forum
asked about how training could be done and most of the time they would
touch a lot of bugs in the software.  There is a great need to
continue the support of SphinxTrain and increase its quality. 

We hope that this release will

1, fix significant amount of bugs of the code and the scripts. 
2, improve the quality of the code and scripts 
3, extend the functionalities of SphinxTrain. 
4, ensure SphinxTrain is fully regression tested. 

New Features:
(r3566, 3567, 3731): mllr_solve, could be used to find the regression matrix
for MLLR. 
(r498, r714): Added the option "-input_endian" so that one can specify
byte.  It follows the the compilation definition.  
order of raw data as different from the machine native byte order.

r580: Prints out links to log files in the html file generated during
training, and creates possibly missing directories.

Bugs:
r444: norm was requesting some parameters that it doesn't really need,
like feature string definition and size of input vector. Removed the
request.
r496: add checking for wave2feat 
r716: When transcription is empty, corpus.c wouldn't show abnormal
behaviour now
r816:819: Memory problems in wave2feat. 

General Code Architecture Related:
(r 58 to 82): Initial checkin by Prof. Alan Black.  That includes
basic Makefiles, RickyH's perl scripts. 
(r 83 to 162): Miscellaneous fixes of Ricky's perl script. 

Developers:
r579: check for return values in read/write. 

Pre-release notes. 
^^^^^^^^^^^^^^^^^
As at 2004-09-17, there was no official release of SphinxTrain by CMU
or by the developers of Sphinx.  Major reason is that SphinxTrain is
in many senses not stable.  

The developers of Sphinx endeavor to release SphinxTrain.  However,
lack of time is always a big constraint for Sphinx's developers.
Before we can get SphinxTrain release, this release notes will used to
record the change made in SphinxTrain correspond to a particular
release of sphinx 3.x.  We also note down the corresponding timestamp.

2004-09-17 Arthur Chan (archan@cs.cmu.edu) at Carnegie Mellon Unversity. 
	SphinxTrian timestamp: SPHINX3_5_CMU_INTERNAL_RELEASE
	correspond to sphinx 3.5 release
	Changes since the last record.
	-The feature extraction libraries of SphinxTrain is now
	EXACTLY the same as sphinx3. 
	-New tools introduced:
	1, mllr_solve, given the EM posterior probabilities of the
	mean and regression class definition. This routine estimate
	regression matrices for each class. 
	2, mllr_transform, given a matrix, this tools applies the
	transformation to the mean vector of the models. 
	-The command line interface for 35 out of 37
	SphinxTrain's tools are unified. Only wave2feat and QUICKCOUNT
	are not yet bind to that rule. 

