rel40.me

speech signal process tools

.nf
.ta 1.5i
  \fIsymsize\fP \-	return the size of a parameter file symbol
  \fIsymlist\fP \-	get list of parameter names and return number of parameters
  \fIsymdefinite\fP \-	indicates whether a run-time parameter prompt will occur
  \fIsymchoices\fP \-	get list of discrete choices for a parameter's value
  \fIsymrange\fP \-	get the value range for a parameter 
  \fIsymprompt\fP \-	returns the prompt string for a parameter
  \fIgetsymdef_X\fP \-	get default parameter value of type X from the symbol table
  \fIfputsym_X\fP \-	put a parameter value of type X into an arbitrary file

.fi
.sp .5
.sh 2 "Other Changes and Additions" 
.np
\fIUpdate_waves_gen\fP was generalized to handle FEA_SD.  The
\fIstart\fP and \fIstep\fP parameters were changed from INT to FLOAT
in order to handle some important special cases (e.g., less than one
output record per input record).  Also, the function now handles
correctly the case in which the generic header item "start_time" is 
is vector.  
.np
Added to support processing of generic header items:
.sp .5
.nf
.ta 1.5i
  \fIget_genhd_val\fP \-	get the value of a numeric generic header item
.fi
.sp .5
\fIget_genhd_val\fP differs from the existing generic header get
routines in returning a value rather than a pointer.  Furthermore, the
value is always returned as a double.
.np
Added to support header processing: 
.sp .5
.nf
.ta 1.5i
  \fIinhibit_hdr_date\fP \-	inhibit updating the header field date
.fi
.sp .5
This routine is used in \fIcomment\fP and \fIaddgen\fP to leave
unchanged the original header creation data.  
.np
Added to support FEA file processing:
.sp .5
.nf
.ta 1.5i
  \fIfea_compat\fP \-	check compatibility of field definitions in FEA headers	
.fi
.np
Added to support ASCII conversions: 
.sp .5
.nf
.ta 1.5i
  \fIatoarray\fP \-	convert ASCII file to data array of specified type
.fi
.sp .5
This function generalizes \fIatoarrayf\fP, which is hard-wired to
produce float arrays.  The new function supports any type (including
complex types).
.np
Added to support general type conversions: 
.sp .5
.nh
.ta 1.5i
  \fItype_convert\fP \-	convert array from one numeric type to another
.fi
.sp .5
This is an elaborate function,  It supports all data types (including 
complex types), and it includes a user-definable action on clipping.  
.np
Miscellaneous additions: 
.sp .5
.nf
.ta 1.5i
  \fIzero_fill\fP \-	fill a newly allocated or existing numeric array with zeros
  \fIdel_fea_fld\fP \-	delete a field from an ESPS Feature File Header
  \fItypesiz\fP \-	return the size of an element of a given ESPS data type
  \fIcover_type\fP \-	returns data type that can contain any data from two input types
.fi
.sp .5
Note that these work with all data types (including complex).  
.np
Added to support FFT processing:
.sp .5
.ta 1.5i
.nf
  \fIget_fftd_inv\fP \-	 compute inverse FFT of doubles
.sp .5
.np
\fIwindow\fP (3\-\s-1ESPS\s+1) has been extended to include a new
window type, WT_ARB, which permits windowing with an arbitary array 
of coefficients.  
.np
The input, output, and state arrays were changed from FLOAT to DOUBLE
in \fIblock_filter\fP (3\-\s-1ESPS\s+1) and \fIinterp_filt\fP
(3\-\s-1ESPS\s+1).  
.np
Added to support mu-law and A-law coding: 
.sp .5
.nf
.ta 1.5i
  \fIlinear_to_mu\fP \-	converts 14-bit integer data to 8-bit mu-law compressed data
  \fImu_to_linear\fP \-	converts 8-bit mu-law compressed data to 14-bit data
  \fIlinear_to_a\fP \-	converts 13-bit integer data to 8-bit A-law compressed data
  \fIa_to_linear\fP \-	converts 8-bit A-law compressed data to 13-bit data
.fi
.np
The following functions were added to support maximum entropy (MAXENT)
and minimum relative-entropy (MRE) calculations:
.sp .5
.nf
.ta 1.5i
  \fImin_rel_ent\fP \-	compute MAXENT or MRE probability distributions
  \fIrel_ent\fP \-	compute relative entropy of two probability distributions
.fi
.np
The \fIget_burg\fP functions (computes reflection coefficients using
standard Burg and modified Burg methods) was modified to use doubles
for internal variables.
.sh 1 "MISCELLANEOUS BUG FIXES" 
.lp
Numerous minor changes and bug fixes are included in the new release.
The more important ones are mentioned here.  
.np
The \fB-S\fP on \fIbtosps\fP now works.  
.np
In Version 4.0a, \fItestsd\fP was introducing slight regularities 
in noise sequences.  This has been fixed.  
.np
FEA_SPEC files of type ARB_FIXED are now supported properly by the 
FEA_SPEC support functions and by \fIpsps\fP.  
.np
\fIme_spec\fP now reads the default parameter file \fIparams\fP if the
\fB-P\fP is not used (no file was read in the previous version.).
.np
The man page for \fIaddgen\fP (1\-\s-1ESPS\s+1) was changed to note
that \fIaddgen\fP doesn't change "File header date".  (Also,
\fIaddgen\fP was changed so that the user's name is included in the
comment that documents when \fIaddgen\fP was run.) 
.np
\fIcomment\fP was changed so that the user's name is included in the
the date/time stamp for an added comment.  
.np
\fIbtosps\fP no longer core dumps when the number of samples 
happens to be a multiple of 1024.  
.np
\fIaddsd\fP (1\-\s-1ESPS\s+1) no longer core dumps when writing
comments in the header about input files that happen to have long path
names.
.np
\fIsfconvert\fP (1\-\s-1ESPS\s+1) now checks to make sure that temporary 
files can be written.
.np
The \fIfea_stats\fP (1\-\s-1ESPS\s+1) man page stated that the program
computes the sample deviation, but it actually computes the unbiased
estimate of the population standard deviation.  The man page has been
corrected.
.np
\fIget_fft\fP, \fIget_fft_inv\fP, \fIget_fftd\fP, and \fIget_rfft\fP
were changed so that they no longer repeatedly allocate storage for new
versions of internal tables without freeing the old versions.
\fIget_rfft\fP was fixed so that it no longer gives the wrong answers
for the special case of length-1 transforms.
.np
\fImergefea\fP how handles the \fIstart_time\fP generic in a more
sensible way (the value is taken from the input file when the output
file is first created; if the output file already exists,
\fIstart_time\fP is not changed).  
.sh 1 "FILE COMPATIBILITY"
.lp
ESPS Version 4.0 will read any files written by programs compiled
under previous releases.  Programs compiled from previous previous
releases, however, will not in general read files written by Version
4.0 programs.   In most cases, programs can be converted to 4.0 merely
by compiling them with the ESPS 4.0 libraries.  (If a sampled data
program is to be converted to use FEA_SD, some changes will be
necessary).  
.sh 1 "WHAT MIGHT BREAK"
.lp
We have tried to make changes so that existing scripts (and habits)
will still work with the new release or will be easy to change.  For
example, although the \fB-d\fP option on copysd has been generalized
to use the full standard list of data types (e.g., float, float_cplx,
etc.), the old codes (e.g., "f" for float) will still work.  Some
changes, however, were deemed important enough to make even if they
will break some existing code.  We have tried to keep track of things
in this category and mention them here.
.np
The order of the file parameters in \fIatofilt\fP (1\-\s-1ESPS\s+1)
and \fIzero_pole\fP (1\-\s-1ESPS\s+1) has been changed so that the
order is input file followed by output file (the general convention
for ESPS).  This will break scripts that use the old versions.
.np
The meaning of the \fB-r\fP option in \fIwplay\fP, \fIplay\fP,
\fIsplay\fP, and \fIsdcomp\fP has been changed.  It is now used as a
range option (to be consistent with all other sampled data programs).
The repeat function (old meaning of \fB-r\fP) is now handled by
\fB-R\fP.
.np
The \fB-s\fP (bytest to skip) option for \fIbtosps\fP was changed to
\fB-S\fP.  Also, all calls to \fIbtosps\fP must now supply a comment
(for record keeping); this can be done in the parameter file or by
using the \fB-c\fP or \fB-C\fP option.  
.np
In the interface to \fIupdate_waves_gen\fP (3\-\s-1ESPS\s+1), the
parameters \fIstart\fP and \fIstep\fP were changed from INT to FLOAT.
This will break programs that use \fIupdate_waves_gen\fP if they are
recompiled under ESPS 4.0 without changing the call.  A simple change
will suffice:  just cast the two arguments to FLOAT in the call; e.g.,
.nf
	int start_rec, step;
	...
	update_waves_gen(ihd, ohd, start_rec, step);

should change to 

	int start_rec, step;
	...
	update_waves_gen(ihd, ohd, (float) start_rec, (float) step);
.fi
.lp
The input, output, and state arrays were changed from FLOAT to DOUBLE
in \fIblock_filter\fP (3\-\s-1ESPS\s+1) and \fIinterp_filt\fP
(3\-\s-1ESPS\s+1).  This will break any programs that use them; change
the calling arguments to double and recompile.  
.sh 1 "COMPUTING ARBITRARY EXPRESSIONS FROM FIELDS IN TWO FEA FILES"
.lp
The program \fIselect\fP (1\-\s-1ESPS\s+1) offers powerful facilities
for computing expressions on fields within FEA records.  We are often
asked about computing expressions that involve fields from more than
one FEA file.  In many cases, the desired capability can be achieved
by combining \fImergefea\fP (1\-\s-1ESPS\s+1) with \fIselect\fP.  The
basic procedure is to use \fImergefea\fP to produce a single FEA file
with fields from a variety of source files and then to run
\fIselect\fP on the combined file.  Here's a small example: Suppose
fea1 is a FEA file containing (among other things) a vector fld1 and a
scalar field fld2.  And suppose fea2 contains a scalar field fld3.
Then the following commands will compute the mean of fld1 and add it to
the log of the square root of fld2 divided by the product of fld3
times the first element of fld1:
.nf
.ft CW

  %mergefea fea1 fea2 fea3
  %select -e "mean(fld1) + log(sqrt(fld2/(fld3*fld1[0])))" fea3 > output

.fi
.ft LR
The results are stored (in ASCII) in the file output.  
.sh 1 "Known Problems"
.np
On Convex systems, with our X programs (and perhaps others?), it is
necessary to use the Internet address of the X server when setting the
\fBDISPLAY\fR environment variable.   For example, instead of:
.nf
.ft CW
% setenv DISPLAY fred:0.0
.fi
.ft R
.br
It is necessary to use the command:
.nf
.ft CW
% setenv DISPLAY 134.45.232.8:0.0
.fi
.ft R
.br
Where 134.45.232.8 is the Internet address of host fred.
.np
On Convex systems, depending on the X server, the \fIxmcd\fR program
might fail and result in a white plot on a white background.
(\fIxmcd\fR is called by the plot programs to do the X Window drawing.)
There is an alternate binary, \fI/usr/esps/bin/xmcd.old\fR, that should
be renamed to \fIxmcd\fR if the standard \fIxmcd\fR fails.   
.np
On Masscomp/Concurrent systems it may be necessary to set the \fIDISPLAY\fR
environment variable to the full hostname of your system, rather than
the generic unix:0.0.   For example, if the network hostname of your
system is fred, then you may have to:
.nf
.ft CW
% setenv DISPLAY fred:0.0
.fi
.ft R
.np
On systems not running a network, in order the for the license server
software to operate you need to set the \fBELM_HOST\fR variable to
\fIlocalhost\fR, instead of the network hostname.   Also in this case,
you may need to set the X Window \fBDISPLAY\fR variable to localhost
also (\fIe.g.\fR localhost:0.0).