//-----------------------------------------------------------------------------------------------------------------------------------------------------------
int IvExtractorEigenDecomposition(Config& config)
{
String inputWorldFilename = config.getParam("inputWorldFilename");
// label for selected frames - Only the frames associated with this label, in the label files, will be used
bool fixedLabelSelectedFrame=true;
String labelSelectedFrames;
if (config.existsParam("useIdForSelectedFrame")) // the ID of each speaker is used as labelSelectedFrame ?
fixedLabelSelectedFrame=(config.getParam("useIdForSelectedFrame").toBool()==false);
if (fixedLabelSelectedFrame) // the label is decided by the command line and is unique for the run
labelSelectedFrames=config.getParam("labelSelectedFrames");
try{
MixtureServer ms(config);
if (verbose) cout << "(IvExtractor) Approximate i-vector by using Eigen Decomposition"<<endl;
if (verbose) cout << "(IvExtractor) TotalVariability - Load world model [" << inputWorldFilename<<"]"<<endl;
MixtureGD& world = ms.loadMixtureGD(inputWorldFilename);
//Read the NDX file
String ndxFilename = config.getParam("targetIdList");
//Remove the first element of each line which is the model name
XList tmpFileList(ndxFilename);
XList fileList;
for(unsigned long ll=0;ll<tmpFileList.getLineCount();ll++){
fileList.addLine();
for(unsigned long i=1;i<tmpFileList.getLine()->getElementCount();i++){
fileList.getLine(fileList.getLineCount()-1).addElement(tmpFileList.getLine(ll).getElement(i));
}
}
//Create and initialise the accumulator
TVAcc tvAcc(fileList, config);
Matrix<double> Q(tvAcc.getRankT(),tvAcc.getRankT());
Matrix<double> D(tvAcc.getNDistrib(),tvAcc.getRankT());
if(config.existsParam("loadEigenDecompositionParam") && config.getParam("loadEigenDecompositionParam").toBool()){ // Load normalized T matrix and weighted Covariance matrix if pre-computed
//Load TotalVariability matrix
String normTFilename = config.getParam("totalVariabilityMatrix") + "_norm";
tvAcc.loadT(normTFilename, config);
//Load D and Q matrices
String dFilename = config.getParam("matrixFilesPath") + config.getParam("totalVariabilityMatrix") + "_EigDec_D" + config.getParam("loadMatrixFilesExtension");
D.load(dFilename,config);
String qFilename = config.getParam("matrixFilesPath") + config.getParam("totalVariabilityMatrix") + "_EigDec_Q" + config.getParam("loadMatrixFilesExtension");
Q.load(qFilename,config);
}
else{
//Load TotalVariability matrix
tvAcc.loadT(config.getParam("totalVariabilityMatrix"), config);
// Normalize matrix T
tvAcc.normTMatrix();
// Compute weighted co-variance matrix by using UBM weight coefficients
DoubleSquareMatrix W(tvAcc.getRankT());
W.setAllValues(0.0);
tvAcc.getWeightedCov(W,world.getTabWeight(),config);
// Eigen Decomposition of W to get Q
Matrix<double> tmpW(W);
Q.setAllValues(0.0);
tvAcc.computeEigenProblem(tmpW,Q,tvAcc.getRankT(),config);
// Compute D matrices (approximation of Tc'Tc matrices)
D.setAllValues(0.0);
tvAcc.approximateTcTc(D,Q,config);
}
//Load the statistics from files or compute statistics for all segments at once
if((config.existsParam("loadAccs")) && config.getParam("loadAccs").toBool()){ //load pre-computed statistics
cout<<" (IvExtractor) Load Accumulators"<<endl;
tvAcc.loadN(config);
tvAcc.loadF_X(config);
}
else{ //Compute statistics if they don't exists
tvAcc.computeAndAccumulateTVStat(config);
tvAcc.saveAccs(config);
}
// Then load the meanEstimate computed by minDiv if required
DoubleVector meanEstimate = tvAcc.getUbmMeans();
if(config.existsParam("minDivergence")&& config.getParam("minDivergence").toBool()){
String minDivName = config.getParam("matrixFilesPath") + config.getParam("meanEstimate") + config.getParam("loadMatrixFilesExtension");
Matrix<double> tmpMean(minDivName,config);
for(unsigned long i=0;i<meanEstimate.size();i++){
meanEstimate[i] = tmpMean(0,i);
}
}
//Update the mean Estimate
cout<<" (IvExtractor) Load Mean Estimate"<<endl;
tvAcc.loadMeanEstimate(meanEstimate);
//Substract mean from the statistics and normalize co-variance
tvAcc.normStatistics(config);
// Estimate I-Vectors
tvAcc.estimateWEigenDecomposition(D,Q,config);
cout<<"--------- save IV by File --------"<<endl;
tvAcc.saveWbyFile(config);
cout<<"--------- end of process --------"<<endl;
} // fin try
catch (Exception& e) {cout << e.toString().c_str() << endl;}
return 0;
}
//-----------------------------------------------------------------------------------------------------------------------------------------------------------
int IvExtractor(Config& config)
{
String inputWorldFilename = config.getParam("inputWorldFilename");
// label for selected frames - Only the frames associated with this label, in the label files, will be used
bool fixedLabelSelectedFrame=true;
String labelSelectedFrames;
if (config.existsParam("useIdForSelectedFrame")) // the ID of each speaker is used as labelSelectedFrame ?
fixedLabelSelectedFrame=(config.getParam("useIdForSelectedFrame").toBool()==false);
if (fixedLabelSelectedFrame) // the label is decided by the command line and is unique for the run
labelSelectedFrames=config.getParam("labelSelectedFrames");
try{
MixtureServer ms(config);
if (verbose) cout << "(IvExtractor) TotalVariability - Load world model [" << inputWorldFilename<<"]"<<endl;
MixtureGD& world = ms.loadMixtureGD(inputWorldFilename);
//Load the statistics from files or compute statistics for all segments at once
//Read the NDX file
String ndxFilename = config.getParam("targetIdList");
//Remove the first element of each line which is the model name
XList tmpFileList(ndxFilename);
XList fileList;
for(unsigned long ll=0;ll<tmpFileList.getLineCount();ll++){
fileList.addLine();
for(unsigned long i=1;i<tmpFileList.getLine()->getElementCount();i++){
fileList.getLine(fileList.getLineCount()-1).addElement(tmpFileList.getLine(ll).getElement(i));
}
}
//Create and initialise the accumulator
TVAcc tvAcc(fileList, config);
//Load TotalVariability matrix
tvAcc.loadT(config.getParam("totalVariabilityMatrix"), config);
//Statistics
if((config.existsParam("loadAccs")) && config.getParam("loadAccs").toBool()){ //load pre-computed statistics
cout<<" (IvExtractor) Load Accumulators"<<endl;
tvAcc.loadN(config);
tvAcc.loadF_X(config);
}
else{ //Compute statistics if they don't exists
tvAcc.computeAndAccumulateTVStat(config);
tvAcc.saveAccs(config);
}
// Then load the meanEstimate computed by minDiv if required
DoubleVector meanEstimate = tvAcc.getUbmMeans();
if(config.existsParam("minDivergence")&& config.getParam("minDivergence").toBool()){
String minDivName = config.getParam("matrixFilesPath") + config.getParam("meanEstimate") + config.getParam("loadMatrixFilesExtension");
Matrix<double> tmpMean(minDivName,config);
for(unsigned long i=0;i<meanEstimate.size();i++){
meanEstimate[i] = tmpMean(0,i);
}
}
//Update the mean Estimate
cout<<" (IvExtractor) Load Mean Estimate"<<endl;
tvAcc.loadMeanEstimate(meanEstimate);
//Substract mean from the statistics
tvAcc.substractM(config);
//Compute vEvT for each session
tvAcc.estimateTETt(config);
// Estimate I-Vectors
tvAcc.estimateW(config);
cout<<"--------- save IV by File --------"<<endl;
tvAcc.saveWbyFile(config);
cout<<"--------- end of process --------"<<endl;
} // fin try
catch (Exception& e) {cout << e.toString().c_str() << endl;}
return 0;
}
