www.pudn.com > voiceprocessingtoolbox.rar > dhmmEval.m

function [maxLogProb, dpPath, dpTable] = dhmmEval(initPI, A, B, O) 
% dhmmEval: Evaluation of DHMM 
%	Usage: [maxLogProb, dpPath, dpTable] = viterbiDecoding(initPI, A, B, O) 
%		initPI: Initial state log probability, 1 x stateNum 
%		A: Transition log probability, stateNum x stateNum 
%		B: State log probability, symbolNum x stateNum 
%		O: Observation sequence of this utterance, frameNum x 1 
%		maxLogProb: Maximum log probability of the optimal path 
%		dpPath: optimal path with size frameNum x 1 
 
frameNum = length(O); 
stateNum = length(A);  
dpTable = -inf*ones(frameNum, stateNum); 
 
if (stateNum>frameNum); error('Number of frames is less than the number of states!'); end 
 
% ====== Fill the first row (matrix view) 
dpTable(1,:)=initPI+B(O(1),:); 
% ====== Fill the first column (matrix view) 
for i=2:frameNum 
	... 
end 
% ====== Fill each row (matrix view) 
for i=2:frameNum 
	for j=2:stateNum 
		... 
	end 
end 
 
% ====== Back track to find the optimum path 
...