ana.1

speech signal process tools

.\" %W% %G% EPI
.TH ANA 1\-SPS %G%
.ds ]W "\fI\s+4\ze\h'0.05'e\s-4\v'-0.4m'\fP\(*p\v'0.4m'\ Entropic Processing, Inc.
.SH NAME
ana \- Performs pitch-synchronous or "vanilla" analysis
.SH SYNOPSIS
.B ana
[
.B \-d
] [
.BI \-x " debug_level"
] [
.BI \-p " param_file"
]
.I [sdfile] pitchfile anafile
.SH DESCRIPTION
.PP
.I Ana
takes as input an SPS speech file and an SPS pitch file.
Using the pitch information, 
.I ana
builds analysis frames and computes
the corresponding reflection coefficients and pulse powers 
per frame.
.I Ana
writes this information in an SPS ANA file.
.PP
For "vanilla" analysis, the analysis frames are of constant length,
as specified by the parameters
.I vced_frmlen
-
MAX(
.I order_vcd, order_unvcd
), 
and
no pitch pulse information is written into the file.
For pitch-synchronous analysis, frame size for un-voiced segments is specified
by
.I
uvced_frmlen,
and 
analysis frames are built using an 
integer number of pitch pulses for voiced segments. The frame
lengths, for voiced frames, are made to be as close as
possible to the value specified by 
.I vced_frmlen.
If the number of pulses in an analysis frame exceeds
.I maxpulses, 
or the number of pulse powers exceed 
.I maxraw, 
or the number of residual powers exceeds 
.I maxlpc,
.I ana 
writes the existing records to the output 
file and stops execution with a warning.
.PP
A special case is 
.I maxpulses = 1.
For this,
.I ana
writes a pulse duration, a raw pulse power, a residual power,
and a set of reflection coefficients for each 
pulse (an unvoiced frame is considered a pulse, as well as a single opening
and closing of the vocal cords during voiced speech). So every ANA record
contains only one pulse duration and set of powers per record.
.PP
An SPS pitch file contains records holding the pitch locations and the length
of the pitch pulse.  The type of the segment (voiced or unvoiced) can be
inferred from the previous information.  The output of 
.I ana,
an SPS ANA file,
contains records holding the location of the frame beginning, the
frame length, pulse duration(s), signal power per pulse, residual power per
frame,
and either
.I order_vcd 
or
.I order_unvcd
reflection coefficients, depending on the
type of frame.
.PP
.SH SPS PARAMETERS
The following parameters are read from the parameter file:
.IP
.sp
.I  "psynch - string"
.IP
"YES", "Y", "y", and "yes" indicates pitch synchronous analysis;
"NO", "N", "no", "n" indicates vanilla analysis.
.sp
.I "start - integer"
.IP
This is the starting point in the sampled data file to begin
analysis. 
.sp
.I "nan - integer"
.IP
This is the number of points (starting at start) to analyze.
.sp
.I "matsiz - integer"
.IP
Useful only for structured covariance and Vector Burg methods. matsiz specifies
the size of the computed matrix.
.sp
.I "win_type - string"
.IP
If it is "HAMMING", a Hamming data window is applied;
if it is "NONE", no data window is applied.
.sp
.I "pre_num - vector"
.IP
A three element vector indicating the numerator of a pole-zero filter - 
for example {1.0, -0.875, 0.0 }.
.sp
.I "pre_den - vector"
.IP
A three element vector indicating the denominator of a pole-zero filter - 
for example {1.0, -0.0, 0.0 }.
.sp
.I "a_method - string"
.IP
Five strings are defined:
"AUTO" - autocorrelation, "BURG" - Burg, "COV" - covariance, "MBURG" - modified
Burg, "STRCOV" - structured covariance, "VBURG" - vector Burg.
.sp
.I  "order_vcd - integer"
.IP
The analysis filter order to use for voiced frames.
.sp
.I  "order_unvcd - integer"
.IP
The analysis filter order to use for unvoiced frames.
.sp
.I "p_offset - integer"
.IP
How many points before the beginning of the first pulse position
to place the analysis window.
.sp
.I  "dcrem - string"
.IP
"YES", "Y", "y", and "yes" indicates dc removal;
"NO", "N", "no", "n" indicates no dc removal.
.sp
.I  "maxpulses - integer"
.IP
How many spaces to allocate for pulse lengths in an analysis frame.
.sp
.I  "maxlpc - integer"
.IP
How many spaces to allocate for raw pulse power values.
.sp
.I  "maxraw - integer"
.IP
How many spaces to allocate for LPC error powers.
.sp
.I "sinc_flg - string"
.IP
If "YES", "Y", "y", or "yes", a SINC function is applied
before autocorrelation extrapolation;
if "NO", "N", "no", or "n", do not filter with a sinc function.
.sp
.I  "uvced_frmlen - integer"
.IP
Analysis window width for unvoiced data segments
.sp
.I  "vced_frmlen - integer"
.IP
Nominal analysis window width for voiced frames and vanilla analysis
.PP
These parameters are also described in ANA(5-SPS).
.PP
.I Min_pp, max_pp
are read from the pitch file header.
Consistency checking is done with the parameters read from the header of the
pitch file and those read from the header of the speech file.  The following 
parameters are checked:
.I sf, ndrec, start and nan.
.SH SPS COMMON
.PP
If the SPS Common file exists and is younger than the params file 
and the name of the input sample data file matches the \fIfilename\fR
entry in the Common file, or an input sample data file is not given on the
command line the Common file is processed.  
In this case, any of the above parameter values that are in the Common
file, will be used from that file.  Otherwise, they will be used from
the params file.
.PP
If the Common file is processed and an input sample data file is not
given on the comamnd line, it is taken from the SPS Common file (entry
\fIfilename\fR).   If, in this case, there is no \fIfilename\fR entry
and no input filename on the command line the program is terminated with
an error message.
.PP
The following values are written into SPS Common:
.I "start - integer"
.IP
This is the point in the sample data file where analysis began. 
.sp
.I "nan - integer"
.IP
This is the number of points analyzed.
.sp
.I "filename - string"
.IP
This is the name of the sample data file that was processed.
.sp
.I "prog - string"
.IP
This is the name of the program that last changed the Common file
(\fIana\fR in this case).
.SH SPS HEADER
A new file header is created for the 
.I ANA
file.
This new header contains values for all the parameters
listed in the 
.I SPS PARAMETERS
section above and for  
.I src_sf, vers, prog, progdate, 
and 
.I ndrec.
In addition,
the source sampled data file name is written to 
.I refer,
and both source-file headers are added as sources in the output file header.
.SH OPTIONS
The following options are supported:
.TP
.B \-d
option specifies that the voicing decision of the pulse tracker should not
be modified using the additional knowledge obtained from spectrum analysis.
An easy way to remember the meaning of
.B
\-d
option is to think that it is similar to the analysis performed by our dsp
(i.e., rosetta) at the present time.
.TP
.BI \-x " debug_level"
If 
.I debug_level
is positive,
.I ana 
prints various information or debugging messages, on the standard error.
The messages are more verbose when
.B debug_level
is larger.  If
.I debug_level
is 0, no message is printed.
The default is 0.
.TP
.BI \-p " param_file"
uses the parameter file 
.I param_file
rather than the default file 
.I params.
.TP
.SH  FILES
.TP
.I sdfile
Input SPS sampled data file.
.TP
.I pitchfile
SPS pitch file.
.TP
.I anafile
Output SPS ANA file.
.SH SUBROUTINES CALLED
analyze, bestauto, covar, fowrdfilt genburg, getatal, getauto, getburg,
getfburg, getvburg, invsefilt, preemp, rctoc, removedc, smartana, toeplib,
wilson.
.SH FUTURE CHANGES
Modifications will occur when we include silence detection in the
algorithms.
.PP
More sophisticated vocing error detection scheme will be
incorporated.
.PP
Other spectral analysis methods might be included.
.SH SEE ALSO
dsptrack (1), ptrack (1), synt (1), SPS(5), ANA(5).
.SH AUTHOR
S. Shankar Narayan. Modified for SPS by David Burton and John Shore.