man.diff

来自「speech signal process tools」· DIFF 代码 · 共 456 行

.\" Copyright (c) 1990 Entropic Speech, Inc. all rights reserved.
.\" @(#)mespec.1	1.8	03 Apr 1997	ESI
.TH ME_SPEC 1\-ESPS 03 Apr 1997
.ds ]W "\fI\s+4\ze\h'0.05'e\s-4\v'-0.4m'\fP\(*p\v'0.4m'\ Entropic Speech, Inc.
.SH NAME
me_spec \- Generate maximum-entropy spectra from parameter vectors.
.SH SYNOPSIS
.B me_spec
[
.BI \-n " num_freqs"
]
[
.BI \-o " max_order"
]
[
.BI \-r " range"
]
[
.BI \-x " debug_level"
]
[
.BI \-G
]
.br
[
.BI \-P " params"
]
.I input
.I output.spec
.SH DESCRIPTION
.I Me_spec
reads spectral parameters and other information from the input file
and generates a corresponding spectral record file containing
logarithmically scaled (in dB) maximum-entropy spectral power densities.
.PP
The spectral parameters may be reflection coefficients,
autoregressive-filter coefficients, or any of several other types of
parameters;
see
.I spec_rep
in the Parameters section and the documentation for
.RI reps_rc(3\-ESPS)
for a list.
Other required information includes the total power, 
the order of the spectrum, and the sampling frequency.
These may be obtained from the input file, or defaults may be used;
see the descriptions of the parameters
.I max_order,
.I samp_freq_name,
.I samp_freq,
.I power_field,
and
.I power.
FEA_ANA(5\-ESPS) files are suitable as input and are specially provided for,
but any vector field of type
.I float
in any FEA file may be treated as a vector of
spectral parameters; see
.I spec_param_field
and
.I spec_rep
in the Parameters section.
.PP
The spectral values are computed for the specified number of frequencies,
which must be at least 2.
The frequencies are equispaced from 0 to half the sampling frequency.
There is an option to compute the spectrum to a lower order than the number
of coefficients present in the analysis records.
.PP
One output record is written for each input record in the specified range.
If ``\-'' is given for the input file, the standard input is read.
If ``\-'' is given for the output file, the standard output is written to.
The output file can be plotted using the plotspec(1\-ESPS) command.
.SH OPTIONS
The following options are supported.  Values in brackets are defaults.
.TP
.BI \-n " num_freqs" "\fR [513]"
Compute spectral values for this number of frequencies, equispaced from 0
to half the sampling frequency.
Must be at least 2.
.TP
.BI \-o " max_order"
Maximum-entropy spectra of at most this order will be computed,
even when more coefficients are available in the analysis records.
The program converts the spectral parameters to reflection coefficients,
carrying the computation out to the actual order of the parameter set
contained in the record.
In FEA_ANA files, when
.B \-G
has not been specified,
the actual order is given by the header item
.I order_vcd,
for records in which the
.I frame_type
field is VOICED or TRANSITION,
and by
.I order_unvcd
in other records.
In other files, or when
.B \-G
has been specified,
the actual order is assumed to be the length of the field
(or subsequence of a field)
named by the parameter
.I spec_param_field.
After converting the parameters,
the program truncates the sequence of reflection coefficients to length
.I max_order
and computes a maximum_entropy spectrum from the truncated sequence.
.IP
This procedure is
.I not
equivalent to simply truncating the original parameter set,
for example by specifing a shorter field with
.I spec_param_field.
The latter procedure may give incorrect results, depending on the spectral
representation (see
.I spec_rep
in the Parameters section).
For example, the
.I m
autoregressive filter coefficients of order
.I m
are not in general the same as the first
.I m
autoregressive filter coefficients of order
.I n
(where
.IR n\|>\|m ).
.TP
.BI \-r " first" : "last" "\fR [(first in file):(last in file)]"
.TP
.BI \-r " first" :+ "incr"
In the first form, a pair of unsigned integers specifies the range
of records to be processed.
Either
.I first
or
.I last
may be omitted; then the default value is used.
If
.IR last " = " first " + " incr,
the second form (with the plus sign) specifies the same range as the first.
If the specified range contains records not in the file,
.I me_spec
complains and quits.
.TP
.BI \-x " debug_level \fR[0]"
If
.I debug_level
is non-zero,
.I me_spec
prints debugging messages and other information on the standard error output.
.TP
.BI \-G
Even if the input file is a FEA_ANA file,
it is treated like any other FEA file.
Special processing does not take place
for determining filter order from voicing
or computing average power from individual pulse powers and durations
in multipulse FEA_ANA files.
The parameter file must supply values for the parameters
.I spec_param_field,
.I spec_rep,
.I power_field
or
.I power,
and
.I samp_freq_name
or
.I
samp_freq.
The filter order is taken to be the length of the field specified by
.I spec_param_field;
but see the discussion of the
.B \-o
option.
.TP
.BI \-P " params" \fR[params]
The name of the parameter file.
.SH ESPS COMMON
The ESPS common file is not referred to.
.SH ESPS PARAMETERS
.PP
The following parameters are read as required.
.TP
.I num_freqs \- integer
The number of frequencies for which spectral values are computed.
The value in the parameter file is not used
if a value is specified with the command-line option
.B \-n.
The default value and restrictions are as for the argument of the option.
.TP
.I max_order \- integer
An upper bound on the order of the maximum-entropy spectra that will be
computed.
The value in the parameter file is not used
if a value is specified with the command-line option
.B \-o.
See the discussion of
.B \-o
in the Options section for the detailed interpretation of this parameter.
.TP
.I start \- integer
The number of the first record to be processed,
counting the first in the input file as number 1.
The value in the parameter file is not used if a starting value is given
with the
.B \-r
option.
The default value is 1.
.TP
.I nan \- integer
The number of records to be processed.
The value in the parameter file is not used if the
.B \-r
option is specified.
By convention, a value of 0 means that all records from the starting record
through the end of the file are processed.
This also the default behavior in case no value is specified either on the
command line or in the parameter file.
.TP
.I spec_param_field \- string
The field containing the spectral parameters.
This must be explicitly specified in the parameter file
unless the input file is a FEA_ANA file and the
.B \-G
option is not specified.
The default in that case is "spec_param".
This parameter may be simply a field name, or it may be of the form
.IP
.IB field-name "\^[\^" grange "\^]\^" \fR,
.IP
where
.I grange
is a general range specification acceptable to
.IR grange_switch (3\-ESPSu).
The bracketed range specification specifies the indices of a subsequence
of the elements of the named field that are to be used as the spectral
parameters.
.TP
.I spec_rep \- string
A name or abbreviation for the type of spectral parameters.
Acceptable values include
"RC" (reflection coefficients),
"LAR" (log area ratios),
"AUTO" (normalized autocorrelation coefficients),
"AFC" (autoregressive filter coefficients),
"CEP" (cepstrum coefficients),
and
"LSF" (line spectral frequencies).
See the documentation for
.RI reps_rc (3\-ESPSsp).
This parameter must be explicitly specified in the parameter file unless the
input file is a FEA_ANA file and the 
.B \-G
option is not specified.
The default in that case is the contents of the generic header item
.I spec_rep.
.TP
.I power_field \- string
A field containing the total power of the spectrum.
Either this parameter or
.I power
must be explicitly specified in the parameter file
unless the input file is a FEA_ANA file and the
.B \-G
option is not specified.
The default in that case is "raw_power".
.TP
.I power \- float
A numerical value for the total power of the spectrum.
This value is used, and applies to all records,
unless the parameter
.I power_field
is defined and names a field that exists in the input file.
In that case, the contents of the named field are used instead.
.TP
.I samp_freq_name \- string
The name of a generic header item that contains the sampling frequency.
Either this parameter or
.I samp_freq
must be explicitly specified in the parameter file
unless the input file is a FEA_ANA file and the
.B \-G
option is not specified.
The default in that case is "src_sf".
.TP
.I samp_freq \- float
A numerical value for the sampling frequency.
This values is used unless the parameter
.I samp_freq_name
is defined and names a generic header item that exists in the input file.
In that case, the value of the header item takes precedence.
.SH ESPS HEADERS
.PP
The values 
.I common.type,
.I variable.refer,
and
.I hd.fea->fea_type
are read from the header of the input file.  The following 
generic header items may also read if the input is a FEA_ANA file:
.I order_vcd,
.I order_unvcd,
.I src_sf,
.I spec_rep
.I frmlen,
.I start,
.I nan,
and
.I maxpulses.
.PP
The following parameters are written to the header of the FEA_SPEC file:
.I common.tag,
.I common.prog,
.I common.vers,
.I common.progdate,
.I variable.refer,
.I frmlen 
(long),
.I sf 
(float),
.I frame_meth
(CODED),
.I freq_format
(CODED),
.I spec_type
(CODED),
.I contin
(CODED),
.I num_freqs
(LONG).
For input FEA_ANA (5\-\s-1ESPS\s+1) files,
.I order_vcd
(LONG)
and
.I order_unvcd
(LONG)
are added to the 
output file header.
In addition, the input file is added as the source of the output file,
and the command line is added as a comment.  
.PP
The generic header item \fIstart_time\fP is written in the output
file.  The value written is computed by taking the \fIstart_time\fP
value from the header of the input file (or zero, if such a header
item doesn't exist) and adding to it the offset time (from the
beginning of the input file) of the first point or record processed.
If \fIrecord_freq\fP doesn't exist in the input file, it defaults to
1.  Otherwise, it is copied to the output file header.  This item
gives the number of records per second of original data analyzed.
.SH SEE ALSO
\fIrefcof\fP (1\-\s-1ESPS\s+1), \fIplotspec\fP (1\-\s-1ESPS\s+1), 
\fIreps_rc\fP (3\-\s-1ESPSsp\s+1), FEA_SPEC (5\-\s-1ESPS\s+1), ESPS
(5\-\s-1ESPS\s+1), 
.SH DIAGNOSTICS
.PP
me_spec: unknown option: 
.RI \- letter
.br
Usage: me_spec [\-n num_freqs][\-o order][\-r range][\-x debug_level][\-G][\-P params] ana_file spec_file
.br
me_spec: too many file names specified.
.br
me_spec: too few file names specified.
.br
me_spec: number of frequencies
.I n
is less than 2.
.br
me_spec: can't open
.I filename: reason
.br
me_spec: 
.I filename
is not an ESPS file
.br
me_spec: Can't start before beginning of file.
.br
me_spec: Last record precedes first.
.br
me_spec: Name of spec_param field not specified.
.br
me_spec: Unrecognized spectral representation.
.br
me_spec: Spectral representation not specified.
.br
me_spec: Field
.I name
and symbol "power" undefined or not float.
.br
me_spec: Neither field nor default value for power specified.
.br
me_spec: Header item
.I name
and symbol "samp_freq" undefined or not float.
.br
me_spec: No header-item name or default value for sampling frequency.
.br
me_spec: Input field
.I name
undefined.
.br
me_spec: Input field
.I name
undefined or not float.
.br
me_spec: Input field length
.I n
too small.
.br
me_spec: can't allocate memory--input spectral parameters
.br
me_spec: can't allocate memory--lpc filter
.br
me_spec: can't allocate memory--reflection coefficients
.br
me_spec: unknown spectral parameter type
.PP
When
.IR debug_level !=0,
the reflection coefficients, average raw power, location tag,
and inverse filter coefficients
are listed for each record.
.SH BUGS
.PP
The
.I frame_len
field in an input FEA_ANA record, defined as ``number of samples to synthesize
from frame,'' is copied to the
.I frame_len
field in a spectral record, defined as ``number of samples in analysis
window.''
This is invalid if the FEA_ANA file was produced by a program for which
the two are different.
.SH WARNINGS
.PP
In order to produce FEA_SPEC (5\-\s-1ESPS\s+1) that can be overlaid
on FEA_SPEC (5\-\s-1ESPS\s+1) files produced by \fIfft\fP (1\-\s-1ESPS\s+1),
the \fB\-G\fP option should be used. This produces a generic FEA_SPEC (5\-\s-1ESPS\s+1) file without the voicing information.
.SH FUTURE CHANGES
.SH AUTHOR
S. Shankar Narayan.
Made SDS compatible by Joe Buck.
Originally called
.I pltspc,
later
.I fanaspec.
Modified for ESPS and number of frequencies made variable
by Rod Johnson, Entropic Speech, Inc.
Modified for FEA_ANA files by John Shore. 
Additions by Rod Johnson to accept spec_reps other than reflection coefficients
and to accept files other than FEA_ANA files.