function X = textgrid2pws (fname, sr)

% TEXTGRID2PWS : reads a PWS-format TextGrid file into a Matlab PWS structure
%
% X = textgrid2pws([fname '.TextGrid'], sr);   
% X = textgrid2pws([fname '.TextGrid']);          % X.sr will have sample rate set to -1
%                                                sample rate isnt used for anything else
% fname does not include the '.TextGrid' extension.
% 
% See http://people.cs.uchicago.edu/~dinoj/tonerec/ for more details

if nargin==2
  X.sr = sr;
else
  X.sr = -1;
end

TONES = '12345';           % assumed that these are tone labels

z = textread([fname '.TextGrid'],'%s','commentstyle','matlab','headerlines',6,'delimiter','\n','bufsize',4096000);

b=sscanf(z{1},'size =%d');
if (~isnumeric(b) & ~length(b))
  b=sscanf(z{1},'size=%d');
end
if ~isnumeric(b)
  error(sprintf('%s.TextGrid not right PWS format',fname));
end

% b should be 4

X.filename = fname;

i=2;
numP = '';  % number of phrases
while  ((i<=length(z)) & ~isnumeric(numP))
  numP = sscanf(z{i},'intervals: size = %d');
  i=i+1;
end

startphrases = [];
endphrases = [];

if ~isnumeric(numP)
  error (sprintf('%s.TextGrid not right PWS format - cant find size of first (Interval) Tier',fname));
else
  lastnum = 0;  % last phrase number that is not -1
  for n=1:numP
    j = i + 4*n - 3;
    st = sscanf(z{j},'xmin = %f');
    en = sscanf(z{j+1},'xmax = %f');
    num = sscanf(z{j+2},'text = "%d');
    if (num > 0)
      if (num ~= lastnum + 1)
	error(sprintf('In phrase tier, phrase %d follows phrase %d in %s.TextGrid',num, lastnum, fname));
      end
      lastnum = num;
      
      %	X.phrases{num}.startphrase = st;
      %	X.phrases{num}.endphrase = en;
      
      X.phrases{num}.filename = fname;
      X.phrases{num}.words = {};
      X.phrases{num}.startwords = [];
      X.phrases{num}.endwords = [];
      X.phrases{num}.wordnames = {};
      X.phrases{num}.validword = [];

      startphrases = [startphrases st];
      endphrases = [endphrases en];
    end
  end
  
  i = i + 4*numP;       % z{i} should be 'item [2]:'
  
  if ~length(strfind(z{i+2},'ord'))
    error('Second Interval Tier should be called "word" Tier');
  end
  
  %%%%%% now read second (word) tier %%%%%%%%%%%%
  
  curP = 0;    % current phrase number (phrase numbering starts at 1)
  
  numW = '';
  while  ((i<=length(z)) & ~isnumeric(numW))
    numW = sscanf(z{i},'intervals: size = %d');
    i=i+1;
  end
  
  % z{i} should be 'intervals[1] :' now

  prevP = 0;
  curWinP = 0; % number of this word in current phrase
  
  for n=1:numW
      j = i + 4*(n-1) + 1;
      st = sscanf(z{j},'xmin = %f');
      en = sscanf(z{j+1},'xmax = %f');
      wd = sscanf(z{j+2},'text = "%s');
      if ('"' == wd(end))
	wd = wd(1:end-1);                % remember wd could be '0' 
      end

      if ~length(strfind(wd,'SIL'))

	curP = length(find(startphrases <= st));  
	if (curP>prevP)
	  curWinP = 1;
	else
	  curWinP = curWinP + 1;
	end
	
	if (curP && (endphrases(end) >= en))
	  L = length(X.phrases{curP}.wordnames);
	  X.phrases{curP}.startwords(L+1) = st;
	  X.phrases{curP}.endwords(L+1) = en;
	  X.phrases{curP}.wordnames{L+1} = wd;

	  if ((length(wd)==1) && strcmp(wd,'0'))
            X.phrases{curP}.validword(L+1) = 0;
  	    clear ws;
	    ws.name = 'invalid';
            X.phrases{curP}.words{L+1} = ws;
          else
            X.phrases{curP}.validword(L+1) = 1;
  	    clear ws;
	    ws.name = wd;
	    ws.numwinphrase = -1;        % not known yet
	    ws.poswinphrase = curWinP;
	    ws.sylls = strread(wd,'%s','delimiter',TONES);
	    if ~length(ws.sylls)
	      [st en curP]
	      wd
	      error ('ws.sylls is empty');
	    elseif (1==length(ws.sylls{end}))
	      ws.sylls = ws.sylls(1:end-1);
	    end
	  
	    S=length(ws.sylls);
	    p = 0;
	    for s=1:S
	      p = p + length(ws.sylls{s}) + 1;
	      ws.tones(s) = str2num(wd(p));          % assuming tone labels only take one digit!
	      ws.sylls{s} = sprintf('%s%s',ws.sylls{s}, wd(p));
	    end
	    ws.sylls = ws.sylls';
	    ws.startsylls = zeros(1,S);
	    ws.endsylls = zeros(1,S);
	    ws.startsyllsPI = zeros(1,S);
	    ws.endsyllsPI = zeros(1,S);
	    ws.startphones = [];
	    ws.endphones = [];
	    ws.phones = {};
	    X.phrases{curP}.words{L+1} = ws;      % ws is a word structure
          end
	end

	prevP = curP;
      end
    end  % end for n=1:numW 
    
    % do some checks
    
    for p=1:length(startphrases)
      numWinP = length(X.phrases{p}.startwords);
      if (startphrases(p) ~= X.phrases{p}.startwords(1))
	warning('startphrases(%d) = %f isnt equal to X.phrases{%d}.startwords(1) = %f\n',p,startphrases(p),p,X.phrases{p}.startwords(1));
      end
      if (endphrases(p) ~= X.phrases{p}.endwords(end))
	warning('endphrases(%d) = %f isnt equal to X.phrases{%d}.endwords(end) = %f\n',p,endphrases(p),p,X.phrases{p}.endwords(end));
      end
      for w=1:numWinP
	X.phrases{p}.words{w}.numwinphrase = length(X.phrases{p}.words);
      end
    end
end

i = i + 4*numW;       % z{i} should be 'item [2]:'


%%%%%%%%%%%%%%%%% now read third (syllabic) tier %%%%%%%%%%%%%%%%%%%%%%%%

curP = 0;    % current phrase number (phrase numbering starts at 1)
curW = 0;    % current word number (word numbering starts at 1)

numS = '';   % number of syllable intervals
while  ((i<=length(z)) & ~isnumeric(numS))
  numS = sscanf(z{i},'intervals: size = %d');
  i=i+1;
end
    
% z{i} should be 'intervals[1] :' now
    
prevP = 0;
prevW = 0;
curWinP = 0; % number of this word in current phrase
numSinW = 0; % number of this syllable in current word

for n=1:numS
  j = i + 4*(n-1) + 1;
  st = sscanf(z{j},'xmin = %f');
  en = sscanf(z{j+1},'xmax = %f');
  syll = sscanf(z{j+2},'text = "%s');
  if ('"' == syll(end))
    syll = syll(1:end-1);
  end

%  if ~length(strfind(syll,'SIL')) %% but this prevents SIL from being used btwn words of the same phrase

  curP_start = length(find(startphrases <= st));    % current phrase, based on start of current syllable
  curP_end = length(endphrases) - length(find(endphrases >= en)) + 1;

  curP = -1;
  if curP_start == curP_end
    curP = curP_start;
  end

  if (curP == 0)
     error(sprintf('start %f of syllable %s is before start of any phrase',st,syll));
     curP = 1;    % the start of this syllable is just before the start of the phrase...
  end

  if (curP ~= -1)
    curWinP = length(find(X.phrases{curP}.startwords <= st));
    curSinW = [];
    if ~(X.phrases{curP}.validword(curWinP))
      if (~strcmp(syll,'0'))
        warning(sprintf('Assuming that syllable %s (from %f to %f) is in an invalid word',syll,st,en));
      end
      curSinW = 1;
      X.phrases{curP}.words{curWinP}.startsylls(curSinW) = st;
      X.phrases{curP}.words{curWinP}.endsylls(curSinW) = en;
    else
      curSinW = findcell(X.phrases{curP}.words{curWinP}.sylls, syll);
      if (length(curSinW) == 0)
        X.phrases{curP}.words{curWinP}.sylls
        error('cant find %s in X.phrases{%d}.words{%d}.sylls above',syll, curP, curWinP);
      elseif (length(curSinW) == 1)
        X.phrases{curP}.words{curWinP}.startsylls(curSinW) = st;
        X.phrases{curP}.words{curWinP}.endsylls(curSinW) = en;
      elseif (length(curSinW) > 1)      % this can happen e.g. yiyide in 0700_0328
                                        %      e.g. yi1zhong1yi1tai2 in 0730_0043 

	for a = 1:length(curSinW)
  	  cswindex = 1;
          while (st <= X.phrases{curP}.words{curWinP}.startsylls(curSinW(cswindex)))
            cswindex = cswindex + 1;  
	             % it should not ever happen that cswindex > length(curSwinW)
          end
	  
	  X.phrases{curP}.words{curWinP}.startsylls(curSinW(cswindex)) = st;
          X.phrases{curP}.words{curWinP}.endsylls(curSinW(cswindex)) = en;
        end    
      end
    end
    prevP = curP;
  end
  
end


%%%%%%%%%%%%%%%%% now read fourth (phonemic) tier %%%%%%%%%%%%%%%%%%%%%%%%


curP = 0;    % current phrase number (phrase numbering starts at 1)
curW = 0;    % current word number (word numbering starts at 1)
curS = 0;    % current syll number (syll numbering starts at 1)
numN = '';   % number of phoNeme intervals
while  ((i<=length(z)) & ~isnumeric(numN))
  numN = sscanf(z{i},'intervals: size = %d');
  i=i+1;
end
    
% z{i} should be 'intervals[1] :' now
    
curWinP = 0; % number of this word in current phrase
numSinW = 0; % number of this syllable in current word
curNinS = 0; % number of this phoneme in current syllable
curNinW = 0; 
prevWinP = 0;

for n=1:numN
  j = i + 4*(n-1) + 1;
  st = sscanf(z{j},'xmin = %f');
  en = sscanf(z{j+1},'xmax = %f');
  phn = sscanf(z{j+2},'text = "%s');
  if ('"' == phn(end))
    phn = phn(1:end-1);
  end

  curP_start = length(find(startphrases <= st));    % current phrase, based on start of current phoneme
  curP_end = length(endphrases) - length(find(endphrases >= en)) + 1;
  curP = -1;
  if (curP_start == curP_end)
    curP = curP_start;
  end
  
  if (curP == 0)    % if curP_start == 0 == curP_end 
     error(sprintf('start %f of phoneme %s is before start of any phrase',st,phn));
     curP = 1;    % the start of this phoneme is just before the start of the phrase...
  end

  % if curP is -1 then this phoneme is between two phrases (which is fine, no need to do anything)

  if (curP ~= -1)
    curWinP = length(find(X.phrases{curP}.startwords <= st));
    if (X.phrases{curP}.validword(curWinP))
      if (prevWinP == curWinP)
        curNinW = curNinW + 1;
      else
        curNinW = 1;
      end
      X.phrases{curP}.words{curWinP}.startphones(curNinW) = st;
      X.phrases{curP}.words{curWinP}.endphones(curNinW) = en;
      X.phrases{curP}.words{curWinP}.phones{curNinW} = phn;
    end
  else
    curWinP = curWinP+1;
  end
  prevWinP = curWinP;
end


for i=1:length(X.phrases)
  for j=1:length(X.phrases{i}.words)
    if (X.phrases{i}.validword(j))
      S = length(X.phrases{i}.words{j}.startsylls);
      for s=1:S	
%	[i j s]
%	X.phrases{i}.words{j}.phones
%	X.phrases{i}.words{j}.startphones
%	X.phrases{i}.words{j}.startsylls
	
        X.phrases{i}.words{j}.startsyllsPI(s) = find(X.phrases{i}.words{j}.startphones == X.phrases{i}.words{j}.startsylls(s));
        X.phrases{i}.words{j}.endsyllsPI(s) = find(X.phrases{i}.words{j}.endphones == X.phrases{i}.words{j}.endsylls(s));
      end
    end
  end
end