gauss_kl.m

来自「利用HMM的方法的三种语音识别算法」· M 代码 · 共 81 行

function [D] = gauss_kl (mu_q,mu_p,sigma_q,sigma_p)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
%   [D] = gauss_kl (mu_q,mu_p,sigma_q,sigma_p
%
%   computes the divergence 
%                /
%      D(q||p) = | q(x)*log(q(x)/p(x)) dx
%               /
%   between two k-dimensional Gaussian probability
%   densities  given means mu and Covariance Matrices sigam where the
%   Gaussian pdf is given by  
%
%              1                                     T       -1
%   p(x)= ------------------------- exp (-0.5  (x-mu)   Sigma    (x-mu)  )
%         (2*pi)^(d/2) |Sigma|^0.5
%
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

if nargin<4,
  error('Incorrect number of input arguments');
end;

if length(mu_q)~=length(mu_p),
  error('Distributions must have equal dimensions (Means dimension)');
end;
mu_q=mu_q(:);
mu_p=mu_p(:);

if size(sigma_q)~=size(sigma_p),
  error('Distributions must have equal dimensions (Covariance dimension)');
end;

DSq=det(sigma_q);
DSp=det(sigma_p);
K=size(sigma_q,1);

D=log(DSp/DSq)-K+trace(sigma_q*inv(sigma_p))+(mu_q-mu_p)'*inv(sigma_p)*(mu_q-mu_p);
D=D*0.5;