timit.m

TIMIT data base 2. Run following Matlab commands to make sure the installation worked correctly:

%TIMIT interface ver 1.00 (1/1/08)
%Made by Kobi Nistel for SIPL.
%Thanks to
%Yevgeni Litvin.
%Odedd Fridman.
%Exemples:
%  db  = timitdb;  -  loads form defalut path.
%  db2 = timitdb('c:\path'); load from a path.

function [this default_path] = timit(init_dir)

%checks the location of the timitdb directory
data_file_path = which('@timit/timit'); %to be fixed
data_file_path = data_file_path(1:end-7); %to be fixed

%TIMIT path form file: 'default_path.txt'
if (nargin<1)
  default_path = textread([data_file_path ,'\default_path.txt'],'%s'); 
  init_dir = default_path{1};
  for ii=2:length(default_path)
      init_dir = [init_dir ,' ',default_path{ii}];
  end
end

persistent cachedDataBase;
this = databaseclass; 
this.path = init_dir; %sets global path
this.kind = 1; %sets DB as sentence
this.name = 'timit'; %to do:from a file

%checks for exisitence of 'tmdb.mat'
if( ~exist([data_file_path 'tmdb.mat'], 'file'))       
    this = makedb(init_dir,this);            
    save([data_file_path 'tmdb'],'this');
    cachedDataBase = this;
else
     fprintf('Loading %s...\n',this.name);
     if isempty(cachedDataBase)
        load([data_file_path 'tmdb']);
        cachedDataBase = this;
     else
         this =   cachedDataBase;
     end          
end
fprintf('Enteries: %d\n',this.entriesNumber);
end



function this = makedb(init_dir,this)
    fprintf('Generating database file...\n');   
    for train_testC=1:2   %USAGE: Test / Run
        switch train_testC
            case 1, train_test = 'train';
            case 2, train_test = 'test';   
        end            
        for dialectC=1:8 % Dialects
            switch dialectC
             case 1,  dialect = 'dr1';
             case 2,  dialect = 'dr2';
             case 3,  dialect = 'dr3';
             case 4,  dialect = 'dr4';
             case 5,  dialect = 'dr5';
             case 6,  dialect = 'dr6';
             case 7,  dialect = 'dr7';
             case 8, dialect =  'dr8';
            end     
            speaker_dirs = dir([init_dir,'\',train_test,'\',dialect,'\*.']); %speakers
            for sex_speakerC = 1:length(speaker_dirs)
                if( strcmpi(speaker_dirs(sex_speakerC).name(1) , 'F') )
                    sex = 'F';                
                else
                    sex = 'M';
                end
                speaker = speaker_dirs(sex_speakerC).name(2:end); %Cutting the F/M
                direct=[init_dir,'\',train_test,'\',dialect,'\',speaker_dirs(sex_speakerC).name];
                files=dir([direct,'\*.wav']);                        
                for sentenceC = 1:length(files)                  
                    
                    %read on sentence data:
                    this.enteries(end+1).ID = length(this.enteries); %couses an empty cell in db(1)!!
                    this.enteries(end).sentence = files(sentenceC).name(1:end-4); %no exonetion
                    this.enteries(end).usage = train_test;
                    this.enteries(end).dialect = dialect;
                    this.enteries(end).sex = sex;
                    this.enteries(end).speaker = speaker;   
                    
                    %read in sentence data:                                     
                    %words:                    
                    [b,e,name]=textread([direct ,'\',this.enteries(end).sentence ,'.WRD'],'%n %n %s');
                    %a "silence" word thet all not word assieted phonems
                    %can point to.
                    name{end+1}='h#';
                    b(end+1) = 1;
                    e(end+1) = 1;
                    for temp=1:length(name), name{temp} = [name{temp},' ']; end
                    %b = b+1; %the Timit readings starts from zero
                    %e = e+1; %the Timit readings starts from zero
                    this.enteries(end).word.name = char (name);                    
                    this.enteries(end).word.b = b;
                    this.enteries(end).word.e = e;                    
                    this.enteries(end).word.flag =  ones(length(b),1,'int8');                   
                    
                    %phonems:
                    [b,e,name]=textread([direct ,'\',this.enteries(end).sentence ,'.PHN'],'%n %n %s');                   
                    for temp=1:length(name), name{temp} = [name{temp},' ']; end
                    %b = b +1; %the Timit readings starts from zero
                    %e = e +1; %the Timit readings starts from zero
                    this.enteries(end).phoneme.name = char (name);                    
                    this.enteries(end).phoneme.b = b; 
                    this.enteries(end).phoneme.e = e;
                    this.enteries(end).phoneme.flag =  ones(length(b),1,'int8');
                    this.enteries(end).phoneme.from =  length(this.enteries(end).word.flag)*ones(length(b),1,'int8');                                     
                    for pho_in=1:length(b)
                        for jj=1:length(this.enteries(end).word.b)
                          if(this.enteries(end).phoneme.b(pho_in)>=this.enteries(end).word.b(jj)-1)...
                           &&(this.enteries(end).phoneme.e(pho_in)<=this.enteries(end).word.e(jj)+1)
                              this.enteries(end).phoneme.from(pho_in)=jj;
                            continue;
                          end
                        end
                    end
                end
            end            
        end
    end    
    this.enteries = this.enteries(2:end);%becouse the first entery is empty
    this.entriesNumber = length(this.enteries);
end