man.diff

speech signal process tools

.\" Copyright (c) 1986-1990 Entropic Speech, Inc.
.\" Copyright (c) 1991 Entropic Research Laboratory, Inc.; All rights reserved
.\" @(#)lpcana.1	3.12 02 Apr 1997 ESI/ERL
.ds ]W (c) 1991 Entropic Research Laboratory, Inc.
.TH LPCANA 1\-ESPS "02 Apr 1997"
.SH NAME
lpcana \- Performs LPC analysis using a crude pitch-synchronous method
.SH SYNOPSIS
.B lpcana
[
.BI \-P " param_file"
] [
.BI \-r " range"
] [
.B \-p " range"
] [
.BI \-m " anal_method"
] [
.B \-F
] [
.BI \-x " debug_level"
]
.I input_sd output_fea
.SH DESCRIPTION
.PP
Use of this program is not recommended.  For pitch synchronous
analysis, see \fIepochs\fR (1\-\s-1ESPS\s+1) and \fIps_ana\fR (1\-\s-1ESPS\s+1).
.I lpcana
takes as input an ESPS sampled data FEA_SD (5\-\s-1ESPS\s+1) file
.i input_sd,
and it computes LPC filter coefficients and the excitation parameters
using a crude pitch-synchronous method.  The results are stored in the
FEA_ANA (5\-\s-1ESPS\s+1) file \fIoutput_fea\fP.  For each analysis frame of data,
\fIlpcana\fP generates reflection coefficients, power, and voicing
information.  An output FEA_ANA record is generated for each pulse
(several pulses per analysis frame), with enough information to enable
\fIlpcsynt\fP (1\-\s-1ESPS\s+1) to re-synthesize speech.  The nominal
frame size is set by the parameter \fIlpc_frame_size\fP but is
determined adaptively by the program.  If the \fB-F\fP option is 
not used, \fIrefcof\fP performs an additional spectral analysis on 
each pulse within the analysis frame and puts out a different set
of reflection coefficients for each pulse (if \fB-F\fP is used, 
all pulses from the same analysis frame have the same reflection 
coefficients).  
.PP
If "\-" is used in place of 
.I input_sd,
standard input is used.  If "\-" is used in place of
.I output_fea,
standard output is used.
.SH OPTIONS
The following options are supported:
.TP
.BI \-P " param_file"
uses the parameter file
.I param_file
rather than the default file
.I params.
.TP
.BI \-p " range"
Selects a subrange of points to be analyzed. The start and end
points are defined with respect to the original SD file that is the
source of the input FEA_ANA file.  The range is specified using the format 
.I start\-end
or
.IR start:end
or
.IR start:+nan .
Either 
.I start
or 
.I end
may be omitted, in which case the omitted parameter defaults respectively
to the start or end of the input SD file.  
.TP
.BI \-r " range"
\fBr\fP is a synonym for \fBp\fP.
.TP
.BI \-m " anal_method\fR[mburg]"
Specifies the spectrum analysis method.  The default is the modified
Burg method.  Also available are the autocorrelation method ("autoc"),
ovariance method ("cov"), Burg method ("burg"), fast modified Burg
method ("fburg"), stuctured covariance ("strcov" and "strcov1"), and
vector Burg ("vburg", fast approximation to structured covariance.  Of
the two structured covariance methods, the first ("strcov" is
considerably faster and better behaved; "strcov1" is older but
included as it may prove useful on occasion.  The \fB-m\fP option
overrides the value that may be in the parameter file.  The default
applies only if there is no value in the parameter file.
.TP
.B \-F
Specifies use of the framing method of the previous version of
\fIrefcof\fP, in which the reflection coefficients for each pulse in
an analysis frame are determined by a spectral analysis of the entire
frame.  If \fB-F\fP is not used, an additional spectrum analysis is
performed for each pulse.  
.TP
.BI \-x " debug_level \fR[0]\fP"
option specifies that various information or debugging messages be
printed on standard error.
.SH ESPS HEADER
.PP
The generic header item \fIstart_time\fP (type DOUBLE) 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.  Unlike many other ESPS programs, \fIlpcana\fP does
\fInot\fP write the generic header item \fIrecord_freq\fP (used by
\fIwaves\fP+ for time synchronization); this is because the output
records from \fIlpcana\fP result from a variable frame length
analysis.
.PP
The following generic header items are added (in addition to the
standard ones (FEA_ANA\-5):
.IP
.I "p_offset - integer"
.IP
How many points before the beginning of the pulse the spectrum
analysis window is started.
.sp
.IP
.I "dcrem - string"
.IP
Yes means that the DC component was removed before the data was analyzed.
.sp
.IP
.I "psynch - string"
.IP
Yes means that the spectrum analysis is done pitch synchronously.
.sp
.IP
.I "matsiz - integer"
.IP
The size of the autocorrelation matrix that was used in the spectrum
analysis.
.sp
.SH "ESPS PARAMETERS"
.PP
The values of parameters obtained from the parameter file are printed
if the environment variable ESPS_VERBOSE is 3 or greater.  The default
value is 3.
The following parameters are read from the parameter file:
.IP
.I "start - integer"
.IP
This is the starting point in the input file.
Its value is superseded
by a \fB\-p\fR value. The default value is 1.
.sp
.IP
.I "nan - integer"
.IP
This is the number of points to analyze. Its value is superseded
by a \fB\-p\fR value.
.sp
.IP
.I  "lpc_filter_order - integer"
.IP
This is the order of the linear prediction filter that represents the
vocal tract.
The maximum allowable size is 20.
The default value is 10.
.sp
.I  "lpc_frame_size - integer"
.IP
This is the nominal frame size for performing spectral analysis. 
The default value is 160.
.sp
.I  "minimum_pulse_length - integer"
.IP
This specifies the minimum pitch period. The algorithm has been well
tested for the pitch range of 50-400 Hz. Thus a typical value of
this variable is 20, assuming a sampling rate of 8000 Hz.
The default value is 20.
.sp
.I "method - string"
.IP
The spectrum analysis method to use.  The available methods are
autocorrelation ("autoc"), covariance ("covar"), Burg ("burg"),
modified Burg ("mburg"), fast modified Burg method ("fburg"),
stuctured covariance ("strcov" and "strcov1"), and vector Burg
("vburg", a fast approximation to structured covariance.  If no value
is given in the file, the modified Burg method is used by default.
The \fImethod\fP is not read from the parameter file if the command
line option \fB\-m\fP is used. 
.SH ESPS COMMON
If the input
is standard input, COMMON is not read. If COMMON is read and the 
command line input filename does match the filename listed in COMMON
then, the following items are read. If the two filenames do not
match, then no further parameters are read from COMMON.
.IP
.I "start - integer"
.IP
This is the starting point in the input file. Any \fB\-p\fR
option value supersedes the COMMON specified value.
.IP
.I "nan - integer"
.IP
This is the number of points to analyze. A \fB\-p\fR specified value
supersedes the COMMON specified value.
.PP
ESPS Common processing may be disabled by setting the environment variable
USE_ESPS_COMMON to "off".  The default ESPS Common file is .espscom 
in the user's home directory.  This may be overidden by setting
the environment variable ESPSCOM to the desired path.  User feedback of
Common processing is determined by the environment variable ESPS_VERBOSE,
with 0 causing no feedback and increasing levels causing increasingly
detailed feedback.  If ESPS_VERBOSE is not defined, a default value of 3 is
assumed.
.SH COMMENTS
The algorithm has been developed for speech sampled at 8000 Hz. It
works best when the data is recorded using a good microphone. Its
quality is degraded, when working with telephone speech (especially
if carbon microphone is used).
.PP
The speech data is not pre-emphasized prior to LPC analysis, and
the modified Burg method is used for computing the LPC parameters.
.SH FUTURE CHANGES
NONE.
.SH BUGS
If the input data consists of a long string of zeros, the program
bombs.
.SH SEE ALSO
.PP
\fIrefcof\fP (1\-\s-1ESPS\s+1), \fIlpcsynt\fP (1\-\s-1ESPS\s+1), 
FEA_ANA (5\-\s-1ESPS\s+1),
.br
\fIFEA_SD\fP (5\-\s-1ESPS\s+1), 
\fIcompute_rc\fP (3\-\s-1ESPS\s+1sp)
.SH AUTHOR
S. Shankar Narayan.  ESPS 3.0 modification by David Burton.
Modified for \fIcompute_rc\fP (3\-\s-1ESPS\s+1sp) and for single-pulse output by 
John Shore.