rel4.1.me

来自「speech signal process tools」· ME 代码 · 共 808 行 · 第 1/3 页

front-end for \fIacf\fP.
.np
\fIcbkd\fP \- new program that computes distances between codewords in
vector quantization codebooks (FEA_VQ files).
.np
\fIdtw\fP \- new program computes dynamic time warping distance
between two sequences stored as feature file fields.
.np
\fIdtw_rec\fP \- new program that compares test sequences to reference
sequences using a dynamic time warping algorithm.  Comparisons can be
made between sequences of floating point vectors or sequences of VQ
codeword indices.
.np
\fIaddclass\fP \- new program for 
VQ-based classification; \fIaddclass\fP adds 
a vector quantization codebook to a set of codebooks being prepared for
classification computations (each codebook represents one class).
.np
\fIvqdst\fP \- new program for VQ-based classification; \fIvqdst\fP
encodes an input feature with respect to each of a set of vector
quantization codebooks (each codebook represents one class); the
encoding rule is minimum mean square error.
.np
\fIvqclassify\fP \- new program for VQ-based classification; applies a
classification rule to the output of \fIvqdst\fP (initial program implements
the "voting" rule - i.e., it selects the class with the largest number
of encodings.
.np
\fInodiff\fP \ new program to compute Nth order differences 
of an arbitrary field in a FEA file.  
.np
\fIxpz\fP \- new program for interactive filter-design program.
\fIxpz\fP provides mouse operations to add, delete, or move poles and
zeros displayed against the unit circle.  As poles and zeros are
modified, the resulting spectral response or other derivative function
is shown in a separate window.
.np
\fIfftcep\fP \- new program for computing complex FFT cepstrum; will
process multi-channel and complex inputs.  
.np
\fIclip\fP \- new program to apply clipping or center clipping to 
FEA fields.  This is useful in certain VQ thresholding applications, 
among others.  It can also be used from \fIxwaves\fP+ to replot 
data on an expanded scale.  
.np
FEA_FILT (5\-\s-1ESPS\s+1) \- this is a new FEA subtype that is used
for store digital filters.  All filter-related programs
(\fInotch_filt, wmse_filt, iir_filt, filter, fft_filter, filtspec,
toep_solv\fP, etc.) have been modified to use the file type.  In
addition to the filter coefficients, FEA_FILT files contain optional
fields to store the poles and zeros.  The \fIxpz\fP, \fInotch_filt\fP,
and \fIiir_filt\fP design programs programs have been modifed to
output poles and zeros in addition to coefficients.  Conversions to
and from the old file format are available via \fIfea2filt\fP
(1\-\s-1ESPS\s+1) and \fIfilt2fea\fP (1\-\s-1ESPS\s+1).  
.np
\fIxrefcof\fP \- new script that runs \fIrefcof\fP on a single frame
of data after using a pop-up window for parameter prompts; the
analysis results are displayed in two pop-up windows, one containing
the reflection coefficients, and one containing a maximum-entropy
power spectrum computed from these reflection coefficients.  It is
sometimes useful to attach \fIxrefcof\fP to the \fIxwaves\fP+ menu
(e.g., via the menus->"->"menu changes"->"add waveform function"
buttons provided in the default startup environment) for cases in
which the \fIxspectrum\fP attachment gets into computational trouble.
.np
\fIxfft\fP \- new script that runs \fIfft\fP on a single frame of data
after using a pop-up window for parameter prompts; the analysis
results are displayed as a pop-up spectral plot.  It is sometimes
useful to attach \fIxrefcof\fP to the \fIxwaves\fP+ menu (e.g., via
the menus->"->"menu changes"->"add waveform function" buttons provided
in the default startup environment) for cases where \fIxspectrum\fP
proves to be inadequate (e.g., \fIfft\fP (1\-\s-1ESPS\s+1) and hence
\fIxfft\fP will compute arbitrary transforms, limited only by memory
size).
.np
\fIcnvlab\fP \- new user-level program for converting NIST/Sphere
(e.g., TIMIT) label files to \fI(x)waves\fP+ format.  
.np
\fIget_esps_base\fP \- new program for locating the
base of the installed ESPS tree (returns $ESPS_BASE or default
/usr/esps).  Useful for writing position-independent scripts. 
.np
\fIfind_esps_file\fP \- new program for finding an ESPS file within
the ESPS tree, with options for a default search path and an
environment variable that overrides the default.  Like
\fIget_esps_base\fP, \fIfind_esps_file\fP is useful for writing
position-independent scripts.
.np
\fImat2fea\fP/\fIfea2mat\fP \- new programs to convert between 
MATLAB .mat files and ESPS FEA files.  
.np
\fIxloadimage\fP \- new program; an enhanced version of Jim Frost's
public domain \fIxloadimage\fP program for displaying images.  Our
variant allows a "slideshow" to run under control of an external
program, \fInext_slide\fP (also included); we use these in one of the
standard demos.  The sources for \fIxloadimage\fP and \fInext_slide\fP 
are included in $ESPS_BASE/pub.
.np
\fIeinfo\fP \- new program to provide a convenient interface to ESPS
release notes and other documentation.  
.np
\fIedemos\fP \- new X interface to a set of demos that illustrate
ESPS/\fIxwaves\fP+ and X Interface Generation (XIG).  
.np
\fIwsystem\fP \- new program to tell what window system you're running
.np
\fIerlsupport\fP \- new convenience program for e-mailing a support
request 
.np
\fIelpr\fP \- new ESPS utility to print ascii text.  
.sh 2 "Major Program Revisions" 
.np
\fIfilter\fP \- generalized to filter multichannel signals and 
complex signals with real filters.  
.np
\fIrefcof\fP \- modified to provide 4 additional autoregressive 
spectrum analysis methods (references are to the corresponding 
ESPS signal processing library routine): 
.nf

    Fast Modified Burg Method (FBURG) \- \fIget_fburg\fP (3\-\s-1ESPS\s+1)

    Structured Covariance Method (STRCOV and STRCOV1) \- see
    \fIbestauto\fP (3\-\s-1ESPS\s+1), \fIstruct_cov\fP (3\-\s-1ESPS\s+1), and \fIgenburg\fP (3\-\s-1ESPS\s+1)

    Vector Burg Method (VBURG) (fast approximation to structured 
    covariance)  \- see \fIget_vburg\fP (3\-\s-1ESPS\s+1)

.fi
.np
\fIlpcana\fP \- modified to provide an option (\fB-m\fP) for selecting
the analysis method used for computing reflection coefficients.  The
available methods are the same 8 methods available from \fIrefcof\fP
(i.e., from the new library routine \fIcompute_rc\fP
(3\-\s-1ESPS\s+1)), namely: 
.nf

    Autocorrelation method (AUTOC)

    Covariance method (COV)

    Burg method (BURG) 

    Modified Burg method (MBURG) 

    Fast Modified Burg method (FBURG) 

    Two Structured Covariance methods (STRCOV and STRCOV1) 

    Vector Burg method (VBURG) (fast approximation to structured 
    covariance)  

.fi
\fIlpcana\fP was also modified to compute reflection coefficients
separately for each frame (i.e., for each pitch pulse).  The old 
framing method is available as an option (\fB-F\fP).  
.np
\fIauto\fP \- \fB-B\fP supports computation via structured covariance 
rather than lag-product.  
.np
\fIsplay\fP \- (SPARCStation codec output) modified to include fast,
approximate resampling code so that files can be played even if they
are not sampled at 8000 Hz.  (Modifications courtesy of Tom Veatch, 
Univ. of Pa.).  \fIsplay\fP also has a \fB-q\fP (quiet) option to
suppress warning messages.  
.np
\fIfeafunc\fP \- the set of computed functions has been expanded
considerably to: \fIabs\fP (absolute magnitude), \fIarg\fP (phase
angle), \fIatan\fP (arctangent), \fIconj\fP (complex conjugate),
\fIcos\fP (cosine), \fIexp\fP (exponential), \fIim\fP (imaginary
part), \fIlog\fP (natural logarithm), \fIlog10\fP (base 10 logarithm),
\fInone\fP (no change \- the identity function), \fIre\fP (real part),
\fIrecip\fP (reciprocal), \fIsgn\fP (signum; for a complex number z
off the real axis, the value is z/|z|), \fIsin\fP (sine), \fIsqr\fP
(square), \fIsqrt\fP (square root), and \fItan\fP (tangent).
Furthermore, the output field no longer has to be a new field,
different from the fields already present in the input; it can now
replace a field with the same name in the input file, including the
input field.  
changed from "ln" and "log" to "log" and "log10" for consistency with
\fIselect\fP and with C; thus the meaning of "log" is now different.
The parameters "gain_factor" and "add_constant" have been replaced
with "gain_real" and "add_real", and "gain_imag" and "add_imag" are
allowed in case complex coefficients are needed.  The same list of
functions available with feafunc can now be applied to arrays of any
of the real and complex ESPS data types in a C program using the 
new library function \fIarr_func\fP.  
.np
\fIplot3d\fP \- modified for optional startup (\fB-w\fP) in a mode
that sends cursor commands to an \fIxwaves\fP+ display server.  Thus,
a spectrogram can be viewed simultaneously in conventional and 
3D formats, with cursor synchronization.  To see an example of this, 
start \fIxwaves\fP+ in server mode (\fB-s\fP), add \fIplot3d\fP 
to the menu via the menus->"menu changes"->"add image function" 
buttons provided in the default startup environment, bring up 
a spectrogram, and invoke \fIplot3d\fP on a marked segment.  
Other \fIplot3d\fP improvements include visible painting of the 
first plot, a \fB-M\fP for a monochrome plot (useful for screen 
dumps), startup allowed without a named file, visible busy signal, 
and load window defaults to current data.  
.np
\fIsgram\fP \- has a new (\fB-T\fP \fIdesired_frames\fP) to force a
particular number of output records; \fIsgram\fP adjusts
\fIstep_size\fP accordingly.  If the start and end times are decreased
while the \fIdesired_frames\fP is kept constant, \fIsgram\fP in effect
computes a spectrogram in the new region with finer time-domain
resolution.  This permits zooming in without interpolation.  This form
of analysis is popular at TI.
.np
\fIvq\fP \- modified so that the codeword index is written to
output file along with quantized fea field, with optional\fB- i\fP 
output of the codeword index alone. 
.np
\fIimage\fP \- modified to display FEA_SPEC fies in ARB_FIXED format; 
also, \fIimage\fP now works properly on untagged FEA_SPEC files. 
Several \fIimage\fP colormaps are now included in the
$ESPS_BASE/lib/colormaps.  
.np
\fIme_sgram\fP \- modified to use support standard ESPS parameter 
file and Common processing (via \fB-P\fP).   \fIme_sgram\fP inherits
the new spectrum analysis methods of \fIrefcof\fP; the \fB-c\fP and
\fB-i\fP options were added to \fIme_sgram\fP to handle optional
parameters in the case of the structured covariance methods. 
.np
\fIfft_filter\fP \- output data type now matches input data type, and 
computations revised for double precision.  
.np
\fIfft, refcof, sgram\fP \- new \fB-z\fP option to suppress warnings.
.np
\fIme_spec\fP \- no longer warns unnecessarily about \fB-G\fP option. 
.np
\fIfilter\fP and \fIrm_dc\fP \- no longer destroy input if output is 
not specified.  
.np
\fIsfconvert\fP \- more robust with respect to erroneous input.  A
\fB-z\fP was added to adjust the output \fIstart_time\fP for the
filter delay.  This permits better synchonization in \fIxwaves\fP+
displays.  
.np
\fIclassify\fP \- modified to remove errooneous assumption that 
the order of records in the statistics file is the same as the
order of the enumerated field "class".  Thanks to Richard Goldhor 
for this.  
.np
\fIcopysd\fP \- new \fB-S\fP option allows range specification in
seconds.  The "standard" option name for this is \fB-s\fP (lower
case), but that's already in use by \fIcopysps\fP (for scale) and we
didn't want to break existing scripts.  However, if \fB-s\fP is now
used with what looks like a range argument, we print a warning.  
A bug was also fixed that prevented \fIstart\fP and \fInan\fP from
being taken from Common as specified in the man page.  Other
Common-related bugs were also fixed.  
.np
\fIcomment\fP \- new \fB-S\fP option to suppress the user-date-time
stamp. 
.np
\fIcopysps\fP \- new \fB-s\fP option to specify range in seconds. 
.np
\fImux\fP \- \fIstart_time\fP generic is now correct (was always
zero), output now correct for 3 or more files (it wasn't), and the
Pmux parameter file now exists.  
.np
\fIaddfead\fP \- modified to accept ascii as well as binary data
(thanks to Richard Goldhor of AudioFile for this); also, there's a
new \fB-F\fP option to save a skipped header as a foreign header;
.np
\fIbtosps\fP \- added \fB-F\fP to save a skipped header as a 
foreign header. 
.np