gnisdisttd.3

speech signal process tools

.\" Copyright (c) 1987-1990 Entropic Speech, Inc.
.\" Copyright (c) 1997 Entropic Research Laboratory, Inc. All rights reserved.
.\" @(#)gnisdisttd.3	1.7 18 Apr 1997 ESI/ERL
.ds ]W (c) 1997 Entropic Research Laboratory, Inc.
.TH GNIS_DIST_TD 3\-ESPSsp 18 Apr 1997
.SH NAME
gnis_dist_td \- compute the gain-normalized Itakura-Saito distortion between two autoregressive power spectra given time domain information.
.SH SYNOPSIS
.ft B
double gnis_dist_td (autocorrn, autocorrn_err, codeword, size)
.br
double *autocorrn, autocorrn_err, *codeword;
.br
int     size;
.PP
.SH DESCRIPTION
.PP
.I gnis_dist_td
computes the gain-normalized 
Itakura-Saito distortion between two power spectra
that are defined by
.I autocorrn 
and 
.I codeword.
.I gnis_dist_td 
computes the distortion by using equation (10) in [1]
in which 
.I autocorrn 
corresponds to the first spectral argument and 
.I codeword 
corresponds to the second spectral argument.
.I autocorrn 
is a set of equally spaced, time-domain autocorrelations 
starting with the zero lag autocorrelation,
.I autocorrn_err 
is the linear predictive analysis squared error that is associated with
the 
.I autocorrn
set of autocorrelations,
and 
.I codeword
is 
the autocorrelation of a set of autoregressive filter coefficients.
.I size
is the number of 
.I autocorrn
and 
.I codeword
values, and
.I gnis_dist_td 
returns
the computed distortion value.
.SH WARNINGS
.PP
.I gnis_dist_td
ignores the first term (element 0) in the arrays,
.I autocorrn
and
.I codeword.
It assumes the arrays are defined from
.I autocorrn[1]
to
.I autocorrn[size]
and
.I codeword[1]
to
.I codeword[size].
.SH EXAMPLE
.PP
.nf
/* Assume the equally spaced, time-domain autocorrelation,
 * \fIauto_corr\fP,  is  defined  from  element 1  to element  6 and the
 * autocorrelation of the autoregressive filter coefficients,
 * \fIcdwd\fP, is defined from element 0 to element 5  (i.e. \fIsize\fP = 6).
 * Also assume that the linear predictive analysis squared error
 * term \fIac_err\fP has been computed.
 * To compute the Gain-Normalized Itakura-Saito distortion measure between
 * the two power spectras, we would have to call \fIgnis_dist_td\fP with the
 * following arguments:
 */

  GNIS_distortion = gnis_dist_td(auto_corrn, ac_err, cdwd \- 1, 6);
.fi
.SH REFERENCES
[1] J. E. Shore and D. K. Burton, Discrete Utterance Speech Recognition Without
Time Alignment," IEEE Transactions on Information Theory, Vol. IT-29, No. 4, July
1983, pp 473-491
.PP
[2] R. M. GRAY, A. Buzo, A. H. Gray, and Y. Matsuyama, "Distortion Measures for
Speech Processing," IEEE Transactions on Acoustics, Speech, and Signal Processing,
Vol. ASSP-28, No. 4, August 1980, pp 367-376
.SH "SEE ALSO"
.nf
\fIgois_dist_td\fP(3\-ESPSsp), \fIis_dist_td\fP(3\-ESPSsp)
.fi
.SH AUTHOR
Code by A. Virdy, manual page by D. Burton