rel33a.me

来自「speech signal process tools」· ME 代码 · 共 158 行

.lo
.ce 2
.b
ESPS 3.3a Release Notes
.sp
Document version: 1.5 4/26/89
.sh 1 "Introduction"
.lp
This document provides notes relevant to release 3.3a of ESPS.  Please
read this document and the Installation Instructions completely before
installing this release of ESPS.  This note documents changes to ESPS
since the revision 3.3.  Please read the 3.3 release notes first.
.sh 1 "Differences Between 3.3 and 3.3a
.lp
Please duplicate this section and make it available to all current users of
ESPS.  This information is needed by others than just the ESPS
installer.
.sh 1 "Major Changes"
.sh 2 "X Window Support"
.lp
The most significant change to this version of ESPS is the added support
for the X Window System.   The current release supports X Windows on
Masscomp and HP systems.   For Masscomp systems, an X version of
\fImcd\fR is distributed, which we call \fIxmcd\fR.  
\fIXmcd\fR is currently only distributed in binary form for Masscomp
systems, since
it will not compile on a standard Masscomp X system (they do not have X
Toolkit and HP Widgets).   For HP systems, this program is distributed
as source, since it will compile.   \fIXmcd\fR will process a subset of
standard GPS produced by the ESPS plot programs.  It might not process
any GPS output you may produce from non-ESPS programs.  See the manual
page for \fIxmcd\fR for details.
.lp
On a Masscomp system, you can build ESPS for either X Windows alone,
SP-40 graphics alone (old MC Windows system), or both.  See the defines
in the install script and the installation instructions.  If you build
ESPS for both X Windows and SP-40 graphics, then the program \fIimage\fR will
default to X Windows, but can be used with SP-40 graphics by using the
\fB-Tmcd\fR option.   See the \fIimage\fR manual page for details.
The other plot programs (\fIplotsd, plotspec, genplot, mlplot\fR, etc)
will only work directly with X windows.  To use them with SP-40, you
need to use the \fB-Tgps\fR option and pipe this to the Masscomp
\fImcd\fR program.  If ESPS is built for only SP-40 grapics, then all of
the plot programs will work correctly.
.lp
For HP systems, we assume X windows as there is no
alternative graphics environment on HP that we support.
.sh 2 "Other Changes"
.np
Fixed a bug in \fImcplay\fR that caused a large FLOAT buffer to be
allocated even when the input file is SHORTS.
.np
\fIsgram\fR modified: Added
"method" to param file -- can be "wb", "nb", or "other".  If "other",
the other parameters are then checked.  If "wb" or "nb", the standard
defaults apply for these cases, although the defaults can be overrided
by command line options.  The parameter "order" was changed to
"fft_order", as was documented in the man page (i.e., the program was
wrong).  Various standard warnings are suppressed unless -x is used
(this so as not to clutter up the output of various scripts).  New
generics: "sgram_method", "fft_order".
.np
\fIvqdesign.c\fR in library updated.  Fixed error message that said
distortion was decreasing when it is increasing.
.np
\fItofspec\fR modified: default output range changed to
(-64,+50) to avoid
sun \fIwaves+\fR colormap probs; added -o and -i options to give
explicit  control over input and output range; related
parameters and generics changed.
.np 
\fIlib/lisp/esps.el\fR added for GNU emacs  support  of ESPS
programming; this file  contains definitions
of  the  interactive functions \fIeman\fR  and \fIecheck\fR - these facilitate
calling the ESPS programs \fIeman\fR and \fiecheck\fR from within
GNU emacs.
.np
The ESPS Programming Guidelines memo was updated to mention the above
information.
.np
\fIfea_stats\fR modified: When \fIfea_stats -R\fR  was run on a
FEA_SPEC file in BYTE format, or in fact any FEA file with more BYTEs
than SHORTs per record, the wrong amount of memory would get allocated
for a certain array containing BYTE-related data, and consequently
other memory would get trashed.
.np
Sun \fImcd\fR (for suntools) fixed:  The change fixes the problem
that garbage is left in a plot window that is resized while partly covered.
.np
\fIhditem\fR man page clarified.
.np
\fIpwr\fR has been revised power to deal with logs; the
functional changes are:
A -l option was added to produce log of power as output instead of 
power.  If -l is used, the default output field name is
"log_power" 
instead of raw_power.  This, as ususal, can be changed via -f
options
or by the contents of the parameter file.  
If the raw power is less than 10/DBL_MAX, logs aren't computed;
in this case the output is limited to 
log10(10/DBL_MAX).
.np
The -p option has been added to \fIsetrange\fR and the -z option has
been documented.  See the man page for details.
.np
\fIme_sgram\fR has been updated to take a -p option.  See the man page
for details.
.np
The makefile and test script for \fIsfconvert\fR has been fixed.
.np
The parser for parameter files had to be changed to operate correctly on
the Sun 386i.  We believe that the problem is related to 386i C compiler
bugs or library bugs.    The problem showed up when using the functions
\fIgetsym_da\fR and \fIgetsym_fa\fR.
.np
Fixed a bug in \fIfft_filter\fR.  The numerator size variable in a zfunc
was not being initialized to zero when it should have been.  This caused
the test program to fail on Sun 386i.
.np
\fIsetmax\fR bug fix:
there was a generic ("double") being
added to the output file.  This was left over from previous
debugging.  It was removed.
.np
A bug in \fItofspec\fR was fixed
that caused it to fail (i.e., core dump) on input files that are not tagged.
.np
Demos updated: The following files were changed:
README, clean, vqdemo, lpcdemo, Xmakeplots, and Sunmakeplots.
There are two new files: createfiles and refparams.
.sh 1 "Possible Problem with Masscomp C Compiler Optimizer"
.lp
On some Masscomp systems we have seen problems with the \fB-O\fR flag.
In particular, the \fIvqdesign.c\fR and \fIvqencode.c\fR files in the
library might have problems with this.   To see if you have this
problem, change to the \fIvq\fR src directory and run the test script in
there.  To do this, just do \fIsh vq.test\fR.   It will be obvious if
the programs fail, because you will get messages about the vq failing to
converge and distortion increasing when it should be decreasing, etc.
.lp
If this is a problem, then you have to recompile the vq library modules
without the \fB-O\fR flag.  To do this, simply edit the 
install script to remove the \fB-O\fR flag from the define for
\fBOLIBCFLAGS\fR, and then do: \fIinstall lib\fR.   You might have to
\fItouch\fR vqdesign.c and vqencode.c in \fIsrc/lib\fR in order to force
them to recompile with \fImake\fR.
.sh 1 "Possible Sun 386i Compiler Problems"
.lp
There is a bug relating to the \fIprintf\fR  \fB%g\fR format character in
SunOS 4.0.0.   When printing values with this format an incorrect number
is printed in some cases.   This format is used in a number of places in
ESPS programs.   We have talked with Sun about this, and they tell us
that it is fixed in 4.0.1.  We have not been able to confirm this
ourselves as of this release.   If you encounter this problem on your
system, please consult with us and we will help you work around it.