Daniel@0: function gmmmixes = mdn2gmm(mdnmixes)
Daniel@0: %MDN2GMM Converts an MDN mixture data structure to array of GMMs.
Daniel@0: %
Daniel@0: %	Description
Daniel@0: %	GMMMIXES = MDN2GMM(MDNMIXES) takes an MDN mixture data structure
Daniel@0: %	MDNMIXES containing three matrices (for priors, centres and
Daniel@0: %	variances) where each row represents the corresponding parameter
Daniel@0: %	values for a different mixture model  and creates an array of GMMs.
Daniel@0: %	These can then be used with the standard Netlab Gaussian mixture
Daniel@0: %	model functions.
Daniel@0: %
Daniel@0: %	See also
Daniel@0: %	GMM, MDN, MDNFWD
Daniel@0: %
Daniel@0: 
Daniel@0: %	Copyright (c) Ian T Nabney (1996-2001)
Daniel@0: %	David J Evans (1998)
Daniel@0: 
Daniel@0: % Check argument for consistency
Daniel@0: errstring = consist(mdnmixes, 'mdnmixes');
Daniel@0: if ~isempty(errstring)
Daniel@0:   error(errstring);
Daniel@0: end
Daniel@0: 
Daniel@0: nmixes = size(mdnmixes.centres, 1);
Daniel@0: % Construct ndata structures containing the mixture model information.
Daniel@0: % First allocate the memory.
Daniel@0: tempmix = gmm(mdnmixes.dim_target, mdnmixes.ncentres, 'spherical');
Daniel@0: f = fieldnames(tempmix);
Daniel@0: gmmmixes = cell(size(f, 1), 1, nmixes);
Daniel@0: gmmmixes = cell2struct(gmmmixes, f,1);
Daniel@0: 
Daniel@0: % Then fill each structure in turn using gmmunpak.  Assume that spherical
Daniel@0: % covariance structure is used.
Daniel@0: for i = 1:nmixes
Daniel@0:   centres = reshape(mdnmixes.centres(i, :), mdnmixes.dim_target, ...
Daniel@0:     mdnmixes.ncentres)';
Daniel@0:   gmmmixes(i) = gmmunpak(tempmix, [mdnmixes.mixcoeffs(i,:), ...
Daniel@0:       centres(:)', mdnmixes.covars(i,:)]);
Daniel@0: end
Daniel@0: