rel4.2.me

来自「speech signal process tools」· ME 代码 · 共 445 行 · 第 1/2 页

length was the implied value of \fInan\fR: in this case the length of
the range from the starting sample to the end of the file.  That
made it impossible to allocate storage for a frame of data without
first reading the entire file.  We changed the default frame length
to be equal to the transform length (2 to the power \fIorder\fR), since
that seemed like a more reasonable default in any case.  Now \fIfft\fR
will run on pipes with \fInan\fR = 0.
.sp
\fIfftinv\fR \fIstart_time\fR calculation has been fixed.
.sp
\fIfft\fR scaling was wrong when frames overlap.
The scaling of spectral values has been revised to be consistent
with the total-power computation.  (The problem was not so much with
overlapping frames as with a frame length different from the transform length.)
.sp
\fIfftinv\fR needed compensating change when \fIfft\fR scaling fixed.
\fIfftinv\fR was changed to compensate for the scaling change in \fIfft\fR
.sp
\fIfftinv\fR needlessly requires \fIfft_length\fR generic.
Now \fIfftinv\fR determines the fft length from the \fInum_freqs\fR header item,
which is standard in FEA_SPEC files, rather than from the \fIfft_length\fR
item, which is not.  If the \fIfft_length\fR item is present, the program
checks for the condition (\fIfft_length\fR == \fInum_freqs\fR - 1) and prints
a warning in case of inconsistency.
.sp
There is now a \fB-t\fR command-line option that controls the data type
of the output FEA_SD file and overrides the data type implied by
the input_data header item.  Any numeric type is now possible, not
just FLOAT and FLOAT_CPLX.
.sp
Under some conditions, \fIfft\fR was writing a bad value for \fInan\fR in the
common file.  It now writes a more appropriate value.
.sp
In the output of \fIfft\fR, the \fIframe_meth\fR header item is now VARIABLE
instead of FIXED, since a partial last frame is possible, and since
the scaling changes made it more important for
\fIframe_len\fR to indicate the actual data length, exclusive of zero padding.
When input is not a pipe, it would be possible to check whether the
last frame has the same length as the rest and to indicate \fIframe_meth\fR
FIXED in that case; this is not done at present, however.  The \fIfftinv\fR
program now handles \fIframe_meth\fR VARIABLE as well as \fIframe_meth\fR FIXED.
.np
\fIexprompt\fR - fixed a floating point conversion error that resulted
in errors when entering floating values to prompts (for example,
entering 204.8, resulted in 204.800003 being stored).
.np
\fIvqdes\fR and \fIvqdesasc\fR - output files were incorrect in the case
of output going to stdout.  This has been fixed.
.np
\fIselect\fR - added -f option to allow specification of a printf option
to use with the -e option.   Previously, %g was used.  This program is
also linked under the name \fIeselect\fR to avoid a name clash with the
\fIksh\fR built in function called \fIselect\fR.
.np
\fIgetparam\fR - a bug was fixed that caused the program to exit without
taking any action if an ESPS common file does not exist.
.np
\fImergefea\fR - 
.sp
The \fB-u\fR option did not work.  We changed the behavior of the \fB-u\fR
option to be consistent with the man
page and changed the man page to be consistent with what the program does
when \fB-u\fR is not specified.
.sp
Mergefea did not check for consistent \fIrecord_freq\fR values.
The program now writes a warning to stderr if two input files exist,
both define values for \fIrecord_freq\fR, and the values are different.  The
message can be suppressed by specifying \fB-z\fR on the command line.
.sp
The \fImergefea\fR program now preserves the FEA file subtype code, if any,
in its second input file, and there is a new \fB-T\fR option for forcing an
arbitrary subtype code in the output.  
.sp
The program issues a warning
if it is given two input files in which different values of the
\fIrecord_freq\fR header item are defined; there is a command-line 
option to suppress the warning.
.sp
Under some conditions, the \fB-f\fR option
could cause other fields as well as those specified to be copied from
the first input file.  This is fixed.
.sp
The \fB-u\fR option would rename generic header items
from the second input file rather than the first.  This is fixed.
.sp
The \fB-t\fR option was
inconsistent with regard to which input file was used as the source for
tag values. This is fixed.
.sh 1 "NEW PROGRAMS"
.np
\fIe2sphere\fR - new program that converts ESPS sampled data files to 
Sphere (NIST formatted) files;
this program is license-free (i.e., can run on any CPU)
.np
\fIsrecord\fR - new program for recording data from the SparcStation 8
bit codecs.  This program also works with the new Sun audio, but
currently is limited to a single channel.
.np
\fIsplay\fR - this is an all new version of the play program for the
SparcStation 8 bit codec.   This program currently is only single
channel and was written for the single channel codec on the older SparcStations.
But it will work on the new higher quality audio such as the
SparcStation 10, but it is limited to only 8000Hz and single channel.
.sp
A version of \fIsrecord\fR and \fIsplay\fR will be available soon for
the new Sun audio hardware.
.sp
SGI users note:  This release also includes \fIsgplay\fR and
\fIsgrecord\fR.  These are full featured record and play programs using
the SGI Indigo built-in audio.  These programs can handle both channels
at up to 48Khz.
.np
\fIzero_pad\fR - new program for adding records to feature files.  It is
intended to be used with the filter program in computing regressions.
.np
\fIv32play\fR - new dual Surf board/FAB2 play program.  This program
does direct from disk playback with the AT&T VME FAB2 board, and the
AT&T or Heurikon Surf board.   In the case of the Surf board, the
program can play two channels.  This program replaces \fIsurf_play\fR
and \fIwplay\fR.
.np
\fIv32record\fR - record program for AT&T or Heurikon VME Surf board.  This
program does not support the FAB2 board.
.sh 1 "CHANGES TO LIBRARY FUNCTIONS"
.lp
.np
\fImatinv\fR - fixed some numerical problems resulting from use of
floats vs. doubles.
.np
\fIblock_filter2\fR - new function for filtering.  See online manual
page.
.np
\fIinit_fdata\fR - new function for filtering.  See online manual page.
.np
\fIfil_spectrum\fR - a new function to find magnitude/phase response
from filter specifications.
.np
\fIinit_anafea_hd\fR - did not add the generic header item
\fIfilt_dsiz\fR and \fIfilt_nsiz\fR when \fIfilters\fR == YES.  Now it
does.
.np
\fIstrcov_auto\fR - uninitialized variable fixed, which caused some
programs (such as \fIrefcof\fR to fail under certain circumstances).
.np
\fIget_vqfea_rec, put_vqfea_rec\fR - these no longer copy the codebook
entries, since marg_index(3\-ESPS) is used in allo_vqfea_rec(3\-ESPS) to setup the
pointers.
.sh 1 "NEW LIBRARY FUNCTIONS"
.lp
.np
\fIsd_to_sphere\fR - converts ESPS sampled data headers to Sphere
headers and converts ESPS sampled data files to Sphere files.  This
function can be used in non-ESPS programs (i.e. it does not require an
ESPS license to run).
.np
\fIeopen2\fR - semantically the same as \fIeopen\fR(3-ESPS) except that, 
unlike \fIeopen\fR, it does not exit if the file is of the wrong type 
(not an ESPS
file or wrong ESPS subytype); instead, \fIeopen2\fR outputs an error
message, and returns.  We had intended to change the semantics
of \fIeopen\fR to work this way, but it turns out that many existing
programs depend on the old behavior.
.sh 2 "LICENSE MANAGER CHANGES"
.lp
Changes have been made to the way the Entropic license manager system
works.  They are:
.np
\fIwcheckout\fR has been removed
.np
\fIwfree has been removed
.np
If an ESPS license is needed and one is not checked out, an auto checkout 
will be attempted for you.
.np
ESPS programs do not talk to the central license server, but instead
talk to a local daemon.  This reduces network traffic and prevents
network delays from affecting programs.
.np
\fIfreewaves\fR added to release a stuck \fIxwaves\fR license
.lp
Normally, \fIecheckout\fR is run to checkout an ESPS license for a particular
host.    If an ESPS program is run, but a license is not checked out on
that host, then the program will try and checkout a license for you.
You will see a message if this is done - and you should be able to tell
if it fails for some reason.
.lp
ESPS programs do not talk to the central license server, but instead
talk to a local daemon.  This reduces network traffic and prevents
network delays from affecting programs.
.sh 1 "THE PUB DIRECTORY" 
.lp
The directory $ESPS_BASE/pub contains full-source copies of 
certain relevant and useful public domain programs. You are 
free to use them and pass them on in any manner consistent with 
restrictions stated by their authors.  This release 
contains these programs:
.ip \fItpipe\fP
This is a utility that splits a \fIunix\fP pipeline into two
pipelines.  Stated differently, the output of one pipe can provide the
intput to two others.  \fItpipe\fP was written by David B. Rosen of
Boston University.
.ip \fIxgrabsc\fP
A utility that allows you to grab arbitrary portions of the 
screen and output them in Postscript or many other formats.
\fIxgrabsc\fP was written by Bruce Schuchardt.  
.ip \fItek2ps\fR
Written by Michael Fischbein, this program converts tektronix graphics
codes into Postscript.  This is used by some of our plotting programs.
.ip \fIshellforms\fP
This is a utility to build and edit forms.  It was written by
Paul Lew.
.sh 1 "KNOWN PROBLEMS"
.np
\fIexprompt\fR - a bug causes all params in a Common file to be included
in the output parameter file, even if not defined in parameter file 
(but only if Common newer than param_file)
.np
Due to new features develped for filtering, an IIR filter file version less
than 1.87 should be re-designed with the new \fIiir_filt\fR program, which
allows better control in filter characteristics.   These old filter files 
(pre FEA_FILT) files will not work - and will cause some programs to fail.