acf.1

来自「speech signal process tools」· 1 代码 · 共 548 行 · 第 1/2 页

.\" Copyright (c) 1987, 1990 Entropic Speech, Inc.; All rights reserved
.\" @(#)acf.1	1.6 4/4/97 ERL
.TH ACF \1-ESPS 11/10/90
.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"
.nf 
acf \- Transforms a data sequence into a set of acoustic features.
.sp
.SH "SYNOPSIS"
.PP
.B acf 
[
.BI \-P " param"
] [
.BI \-{r,s} " range"
][
.BI \-x " debug_level"
]
.I sd_file
.I fea_file
.sp
.SH "DESCRIPTION"
.PP
\fIAcf\fP takes a single channel FEA_SD file, reads
frames of data,
and produces acoustic features based on each
frame. The following acoustic features are supported:
power, zero crossings, Fast Fourier Transform (FFT), autocorrelation, 
reflection coefficients, 
linear prediction coefficients (LPCs), cepstral coefficients derived from 
LPCs and directly from the FFT, log area ratios, and line spectral frequencies.
Mel-warped cepstral features can be generated using the 
bilinear Z-transform.
Parameters are read by
\fIacf\fP from the default parameter file, \fIparams\fP,
or the file specified by the \fB\-P\fP option. 
.PP
The initial point of the first frame is determined by 
the \fB\-r\fP option or by \fIstart\fP in the parameter file.  
The number of data points in each frame is \fIframe_len\fP and the 
initial points of any subsequent frames follow at equal intervals
.I step.
Thus the 3 cases
.IR step " < " frame_len,
.IR step " = " frame_len,
and
.IR step " > " frame_len,
correspond to overlapping frames, exactly abutted frames,
and frames separated by gaps.
.PP
The number of frames extracted
is the minimum sufficient to cover a specified range of
\fInan\fP points, given \fIframe_len\fP and \fIstep\fP.
.PP
The sampled data in each frame optionally may be windowed using a
window function specified by \fIwindow_type\fP 
in the parameter file.  It is possible to store the windowed data 
frame can be stored in one of the fields in the FEA file.
.PP
The acoustic features used to form the records in 
\fIfea_file\fP are selected by the \fIflag\fP parameters in 
the parameter file.  If a flag is set, e.g. an entry \fB int pwr_flag = 1;\fP
appears in the parameter file, then the corresponding feature, in this case the
power in each frame, is computed and stored in \fIfea_file\fP.  Full description
of the flag parameters appears below.  Features are stored in separate fields 
in the FEA file, which are defined only for the selected features.
.PP
The output file, \fIfea_file\fP, is a tagged FEA file.  If the Fast Fourier Transform  
is to be computed, \fIfea_file\fP is of feature-file subtype FEA_SPEC. The tags in
each record give the starting point of the frame in \fisd_file\fP.
.PP
If \fIsd_file\fP is "\-" then the input is read from the standard input and if
\fIoutput\fP is "\-" then the output is directed to the standard output.
.PP
Note that the menu-based X Windows program \fIxacf\fP(1-ESPS) 
provides an easy to use interface to \fIacf\fP.
.sp
.SH "OPTIONS"
.PP
.TP 
.BI \-P " param"
Uses the parameter file \fIparam\fP rather than the default, 
which is \fIparams\fP.
.TP
.BI \-{r,s} " first":"last"
.TP
.BI \-{r,s} " first\-last"
.TP
.BI \-{r,s} " first" :+ "incr" 
The \fB\-r\fP and \fB\-s\fP options specify the range of data to process.
Only one of the two may be used. If \fB\-r\fP is used, the
arguments are in numbers of samples. If \fB\-s\fP is used, the 
arguments are in seconds.
In the first two forms, a pair of positive numbers specifies the range of
sampled data to analyze.  If 
.IR last " = " first " + " incr,
the third form (with the plus sign) specifies the same range as the
first two forms.  If \fIfirst\fP
is omitted, the default is to begin processing at the first point in the
file.  If \fIlast\fP
is omitted, then the entire file is processed; \fIlast\fP cannot be
omitted if reading from standard input.
If the specified range contains points not in the input file, zeros
are appended. Both forms of the option override the values of 
\fIstart\fP and \fInan\fP
in the parameter file or ESPS Common file.  If these options are not
used, the range is determined from the ESPS Parameter or Common file if the
appropriate parameters are present.  If the \fB\-s\fP option is used,
the field \fIrecord_freq\fP in the input header is used to translate the units
in seconds to numbers of samples. If \fB\-s\fP is used and 
\fIrecord_freq\fP is undefined, or is 0, \fIacf\fP warns and exits.
.TP
.BI \-x " debug_level" " \fR [0]"
A positive value specifies
that debugging output be printed on the standard error output.
Larger values result in more output.
The default is 0, for no output.
.sp
.SH "ESPS PARAMETERS"
.PP
\fIAcf\fP requires that a parameter file be present.  If \fIacf\fP
is called without the \fB\-P\fP option, the file \fIparams\fP must
be present.  If the \fB\-P\fP option is used, the specified file must
be present.  If the appropriate parameter file exists, the 
parameters listed below are read from it .  
.PP
Note that some integer parameters
are flags, e.g. \fIsd_flag\fP, \fIpwr_flag\fP.  These flags determine which
acoustic features are stored in \fIfea_file\fP. If the value assigned to a
flag parameter is \fB1\fP, the associated acoustic field is computed and
stored in a field in the output record. If any value
other than \fB1\fP is assigned to a flag parameter, the acoustic
feature associated with the flag is not stored in \fIfea_file\fP. By
default, no flags are set, so if a flag is not set explicitly in the
parameter file, the associated acoustic feature will not be stored in
\fIfea_file\fp .
.PP
There are default field names for all the acoustic featues. These can
all be changed using the field name parameters, e.g. \fIsd_fname\fP,
\fIpwr_fname\fP. Unless otherwise noted, acoustic features have data type
REAL and dimension 1. 
.PP
The following parameters are read from the appropriate parameter file: 
.TP
.I "sd_field_name \- string"
.IP
Specifies the name of the field in the input file \fIsd_file\fP which
contains the sampled data.  The default field name is "sd".
.TP
.I "units \- string"
.IP
Specifies whether the units of the parameters \fIstart\fP, \fInan\fP,
\fIframe_len\fP, and \fIstep\fP are in seconds or samples.  There
are two valid values for \fIunits\fP: "seconds" or "samples". If
"seconds" is specified, the header item \fIrecord_freq\fP is used
to convert the specified values in seconds to an equivalent number of
samples.  If "seconds" is specified and the field \fIrecord_freq\fP is 
not defined in the input header, or is 0, \fIacf\fP warns and exits.  
The default value is "samples". This parameter is not read if
the \fB\-r\fP or the \fB\-s\fP option is used.
.TP
.I "start \- float"
.IP
The first point in the input sampled data file that is processed.  The
first sample in the file is denoted by 1 if \fIunits="samples"\fP or
by 0 if \fIunits="seconds"\fP.
If it is not in the parameter
(or Common) file, the default is to begin processing at the begining
of the file. 
This parameter is not read if 
the \fB\-r\fP or the \fB\-s\fP option is used.
.TP
.I "nan - float"
.IP
The total amount of data to process.  If 
.I nan
is 0 and the input is not a pipe, the whole file is processed.
If \fInan\fP
is 0 and the input is a pipe, \fIacf\fP warns and 
exits.  If \fInan\fP is not specified, the value 0 is used as its
default.
\fINan\fP
is not read if the \fB\-r\fP or \fB\-s\fP option is used.  
.TP
.I "frame_len \- float"
.IP
The length of each frame.  
A value of 0 indicates that a single frame of length
.I nan
is processed; this is also the default value in case
.I frame_len
is not specified in the parameter file. 
.TP
.I "step \- float"
.IP
Initial points of consecutive frames are separated by this amount.
If the option is omitted and no value is found in the parameter file,
a default equal to
.I frame_len
is used (resulting in exactly abutted frames).  
.TP
.I "preemphasis \- float"
.IP
If \fIpreemphasis\fP is not equal 0.0, the first order preemphasis
filter
.br 
(1 - (\fIpreemphasis\fP) z^(-1)) is applied to the data.  The
preemphasis factor must be between 0 and 1; if not, \fIacf\fP warns
and exits.
.TP
.I "window_type \- string"
.IP
The data window to apply to the data.
If the option is omitted and if no value is found in the parameter file,
the default used is "RECT", for a rectangular window with amplitude one.
Other acceptable values include
"HAMMING", for Hamming, and "TRIANG", for triangular;
see the window(3-ESPS) manual page for the complete list.
.TP
.I "sd_flag \- integer" 
.TP
.I "sd_fname \- string"
.IP
If \fIsd_flag\fP is set to 1, the frame of windowed data is stored 
in the output records of \fIfea_file\fP; by default, this is not done.
If \fIsd_fname\fP is defined in the parameter file,
its value specifies the name of the field in which the windowed data
is stored.  The default field name is "sd".
.TP
.I "pwr_flag \- integer"
.TP
.I "pwr_fname \- string"
.IP
If \fIpwr_flag\fP is set to 1, the power in each windowed frame of data
is computed and stored in \fIfea_file\fP.  The
power is the sum of the squared values of the windowed data divided by
the number of points in the current frame.  The power of each frame is
stored in the field "power", unless \fIpwr_fname\fP is assigned a
different name.
.TP
.I "zc_flag \- integer"
.TP
.I "zc_fname \- string"
.IP 
If \fIzc_flag\fP is set to 1, the zero-crossings in each windowed
frame of data are computed and stored in \fIfea_file\fP.  The
zero-crossings are computed as in the program \fIzcross\fP(1-ESPS).
The number of zero-crossings in each frame is  stored in the field
"zero_crossing", unless 
\fIzc_fname\fP is assigned a different name.
.TP
.I "ac_flag \- integer"
.TP 
.I "ac_fname \- string"
.TP
.I "ac_order \- int"
.IP
If \fIac_flag\fP is set to 1, the sample autocorrelation of the
windowed data 
frame for lags 0,...,\fIac_order\fP is found using \fIget_auto\fP
(3-ESPS).  The data is stored in the field "auto_corr", unless
\fIac_fname\fP is assigned a different name.  The output field has
data type REAL and dimension \fIac_order\fP+1. If \fIac_order\fP=0,
a warning message is printed.
.TP
.I "rc_flag \- integer"
.TP
.I "rc_fname \- string"
.IP
If \fIrc_flag\fP is set to 1, reflection coefficients are
found from the sample autocorrelation of the windowed data frame 
using \fIget_atal\fP(3-ESPS). 
The parameter \fIac_order\fP determines the number of reflection
coefficients which are computed. If \fIac_order\fP=0, a warning