function d=distitar(ar1,ar2,mode)
%DISTITAR calculates the Itakura distance between AR coefficients D=(AR1,AR2,MODE)
%
% Inputs: AR1,AR2     AR coefficient sets to be compared. Each row contains a set of coefficients.
%                     AR1 and AR2 must have the same number of columns.
%
%         MODE        Character string selecting the following options:
%                         'x'  Calculate the full distance matrix from every row of AR1 to every row of AR2
%                         'd'  Calculate only the distance between corresponding rows of AR1 and AR2
%                              The default is 'd' if AR1 and AR2 have the same number of rows otherwise 'x'.
%           
% Output: D           If MODE='d' then D is a column vector with the same number of rows as the shorter of AR1 and AR2.
%                     If MODE='x' then D is a matrix with the same number of rows as AR1 and the same number of columns as AR2'.
%
% If ave() denotes the average over +ve and -ve frequency, the Itakura spectral distance is 
%
%                               log(ave(pf1/pf2)) - ave(log(pf1/pf2))
%
% The Itakura distance is gain-independent, i.e. distitpf(f*pf1,g*pf2) is independent of f and g.
%
% The Itakura distance may be expressed as log(ar2*toeplitz(lpcar2rr(ar1))*ar2') where the ar1 and ar2 polynomials
% have first been normalised by dividing through by their 0'th order coefficients.

% Since the power spectrum is the fourier transform of the autocorrelation, we can calculate
% the average value of p1/p2 by taking the 0'th order term of the convolution of the autocorrelation
% functions associated with p1 and 1/p2. Since 1/p2 corresponds to an FIR filter, this convolution is
% a finite sum even though the autocorrelation function of p1 is infinite in extent.
% The average value of log(pf1) is equal to log(ar1(1)^-2) where ar1(1) is the 0'th order AR coefficient.

% The Itakura distance can also be calculated directly from the power spectra; providing np is large
% enough, the values of d0 and d1 in the following will be very similar:
%
%         np=255; d0=distitar(ar1,ar2); d1=distitpf(lpcar2pf(ar1,np),lpcar2pf(ar2,np))
%

% Ref: A.H.Gray Jr and J.D.Markel, "Distance measures for speech processing", IEEE ASSP-24(5): 380-391, Oct 1976
%      L. Rabiner abd B-H Juang, "Fundamentals of Speech Recognition", Section 4.5, Prentice-Hall 1993, ISBN 0-13-015157-2
%      F. Itakura, "Minimum prediction residual principle applied to speech recognition", IEEE ASSP-23: 62-72, 1975

%      Copyright (C) Mike Brookes 1997
%
%      Last modified Fri Jan  7 08:59:56 2000
%
%   VOICEBOX is a MATLAB toolbox for speech processing. Home page is at
%   http://www.ee.ic.ac.uk/hp/staff/dmb/voicebox/voicebox.html
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%   This program is free software; you can redistribute it and/or modify
%   it under the terms of the GNU General Public License as published by
%   the Free Software Foundation; either version 2 of the License, or
%   (at your option) any later version.
%
%   This program is distributed in the hope that it will be useful,
%   but WITHOUT ANY WARRANTY; without even the implied warranty of
%   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
%   GNU General Public License for more details.
%
%   You can obtain a copy of the GNU General Public License from
%   ftp://prep.ai.mit.edu/pub/gnu/COPYING-2.0 or by writing to
%   Free Software Foundation, Inc.,675 Mass Ave, Cambridge, MA 02139, USA.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

[nf1,p1]=size(ar1);
nf2=size(ar2,1);
m2=lpcar2ra(ar2);
m2(:,1)=0.5*m2(:,1);
if nargin<3 | isempty(mode) mode='0'; end
if any(mode=='d') | (mode~='x' & nf1==nf2)
   nx=min(nf1,nf2);
   d=log(2*sum(lpcar2rr(ar1(1:nx,:)).*m2(1:nx,:),2).*((ar1(1:nx,1)./ar2(1:nx,1)).^2));
else
   d=log(2*lpcar2rr(ar1)*m2'.*((ar1(:,1)*ar2(:,1)'.^(-1)).^2));
end