NormFeatWindowMode.cpp

Go to the documentation of this file.

/*
This file is part of LIA_RAL which is a set of software based on ALIZE
toolkit for speaker recognition. ALIZE toolkit is required to use LIA_RAL.

LIA_RAL project is a development project was initiated by the computer
science laboratory of Avignon / France (Laboratoire Informatique d'Avignon -
LIA) [http://lia.univ-avignon.fr <http://lia.univ-avignon.fr/>]. Then it
was supported by two national projects of the French Research Ministry:
        - TECHNOLANGUE program [http://www.technolangue.net]
        - MISTRAL program [http://mistral.univ-avignon.fr]

LIA_RAL is free software: you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License as
published by the Free Software Foundation, either version 3 of
the License, or any later version.

LIA_RAL is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU Lesser General Public License for more details.

You should have received a copy of the GNU Lesser General Public
License along with LIA_RAL.
If not, see [http://www.gnu.org/licenses/].

The LIA team as well as the LIA_RAL project team wants to highlight the
limits of voice authentication in a forensic context.
The "Person Authentification by Voice: A Need of Caution" paper
proposes a good overview of this point (cf. "Person
Authentification by Voice: A Need of Caution", Bonastre J.F.,
Bimbot F., Boe L.J., Campbell J.P., Douglas D.A., Magrin-
chagnolleau I., Eurospeech 2003, Genova].
The conclusion of the paper of the paper is proposed bellow:
[Currently, it is not possible to completely determine whether the
similarity between two recordings is due to the speaker or to other
factors, especially when: (a) the speaker does not cooperate, (b) there
is no control over recording equipment, (c) recording conditions are not
known, (d) one does not know whether the voice was disguised and, to a
lesser extent, (e) the linguistic content of the message is not
controlled. Caution and judgment must be exercised when applying speaker
recognition techniques, whether human or automatic, to account for these
uncontrolled factors. Under more constrained or calibrated situations,
or as an aid for investigative purposes, judicious application of these
techniques may be suitable, provided they are not considered as infallible.
At the present time, there is no scientific process that enables one to
uniquely characterize a persones voice or to identify with absolute
certainty an individual from his or her voice.]

Copyright (C) 2004-2010
Laboratoire d'informatique d'Avignon [http://lia.univ-avignon.fr]
LIA_RAL admin [alize@univ-avignon.fr]
Jean-Francois Bonastre [jean-francois.bonastre@univ-avignon.fr]
*/

#if !defined(ALIZE_NormFeatWindowMode_cpp)
#define ALIZE_NormFeatWindowMode_cpp

#include <iostream>
#include <fstream>
#include <cstdio>
#include <cassert>
#include <cmath>
#include <liatools.h>
#include "NormFeatWindowMode.h"

using namespace alize;
using namespace std;



//VERSION WITHOUT MEMORY
void computeCMVnorm(Feature &f, DoubleVector &cMean, DoubleVector &cCov ){
        
                
        for (unsigned long c = 0; c < cMean.size(); c++)
        {
            f[c] = (f[c] - cMean[c])/cCov[c];
        }       
}



//INIT CMVNorm on a matrix of windowFrame Features (RefVector)

void loadMeanAndCovParam(RefVector <Feature> &windowFrameAcc, DoubleVector &cMean,DoubleVector &cCov, long &windowDuration )
{
    FrameAccGD local;
    //create local Feature
    Feature f(cMean.size());
    int i = 0;
        
    if(windowFrameAcc.size() == (unsigned long)windowDuration) 
    {
        while(i < windowDuration)
        {
           f = windowFrameAcc[i];
           local.accumulate(f);
           i++;
        }
        cMean = local.getMeanVect();
        cCov  = local.getStdVect();
        local.reset();
        
    }
    else
    {
        cout << "Pb with buffer length" <<endl;
    }
   
}

//Update CMVNorm parameters when adding a new feature in the window (faster execution than loadMeanAndCovParam)

void updateMeanAndCovParam(Feature &f, DoubleVector &cMean,DoubleVector &cCov, long &windowDuration, long &frameCount, long &lookHead )
{       
    //Update window mean & var parameters
        
    //Forget factor Beta, need almost double precision
    double Beta;
    //Do not update mean & var parameters as current feature was already processed   
    if(frameCount < lookHead)
    {
        Beta = 1.0;    
    }
    //Update mean & var parameters for "not seen" feature  
    else
    {
        Beta = (((double)windowDuration - 1) / (double)windowDuration) ;
    }
    
    for (unsigned long i = 0; i < cMean.size(); i++)
    {
        cMean[i] = Beta * cMean[i]  + (1 - Beta)*f[i];
        cCov[i]  = sqrt( cCov[i] * cCov[i] * Beta + (1 - Beta) * (f[i]*f[i])); 
        
    }
    
   
}


//COMPUTE parameters on a windowDuration window, update the window by deleting the 1st frame and adding the current frame
void computeCMVparameters(Feature &f, long &windowDuration, RefVector <Feature> &windowFrameAcc, DoubleVector &cMean, DoubleVector &cCov, long &frameCount  )
{       
        // suppress one frame & add the current frame
        
        loadMeanAndCovParam(windowFrameAcc,cMean,cCov,windowDuration );
        
        windowFrameAcc.removeObject(0);
        windowFrameAcc.addObject(f);
        
        
}


/*
Variance and Mean cepstral normalization

Init : used zero padding + initDelay frames before norm OR init from a reference vector (TO DO)                                                                              
For updating norm statistics a forget factor is used = windowFrame-1/windowFrame

TAKE CARE : works only on voiced feature, doesn't use labels

*/
int normFeatOnlineMode (Config & config) {
        
  String inputFeatureFileName =config.getParam("inputFeatureFilename");          // input feature - could be a simple feature file or a list of filename
  String labelSelectedFrames  =config.getParam("labelSelectedFrames");           // Only the frames from segments with this label  will be used 
 
  long windowDuration = 300, // Default length for window = 3s
       frameCount = 0,
       lookHead   = 0;
  bool  initWithDelay = false;
  
  if (config.existsParam("initWithDelay"))
  { 
      lookHead = config.getParam("initWithDelay").toLong();
      initWithDelay = true;
        
  }

  if(lookHead > windowDuration)
  {
      cout << "BAD INIT DELAY, SET TO WINDOW DURATION"<<endl;
      lookHead = windowDuration;
  }
  
  bool writeAllFeature    = true; // Output a vector for all input vectors (selected and not selected vectors) - DEFAULT=on
  if (config.existsParam("writeAllFeatures"))
  {
      writeAllFeature=config.getParam("writeAllFeatures").toBool();    // Define if all the feature (selected or not) should be written 
  }
  
  if (verbose){
    cout << "NormFeatWindowMode";
    cout << " Window mode, Window Duration["<<windowDuration<<"]"<<endl;
    cout << "var and mean normalisation"<< endl; 
  }
  double frameLength = 0.01; // length of a frame ins, by default 10 ms
  
  if (config.existsParam("frameLength")) frameLength=config.getParam("frameLength").toDouble();
 
  
        
  RefVector <Feature> memFrameAcc;
  RefVector <Feature> windowFrameAcc;
  
  XLine inputFeatureFileNameList;                                                // The (feature) input filename list
  if ( inputFeatureFileName.endsWith(".lst")){                                   // If the file parameter is the name of a XList file
        XList inputFileNameXList(inputFeatureFileName,config);                   // Read the filename list file
        inputFeatureFileNameList=inputFileNameXList.getAllElements();            // And put the filename in a list if the file is a list of feature filenames
  }
  else {                                                                         // It was a simple feature file and not a filename list
    inputFeatureFileNameList.addElement(inputFeatureFileName);                   // add the filename in the list
  }
  try{
        String *file;
        
          // Loop on each feature file
        while ((file=inputFeatureFileNameList.getElement())!= NULL)
        {                         
            String & featureFileName=(*file);
            FeatureFileReader fr(featureFileName, config);
                
            if (!(config.existsParam("featureFlags")))
                config.setParam("featureFlags",fr.getFeatureFlags().getString());
            
            if (config.existsParam("windowDuration"))
            {     
                windowDuration = config.getParam("windowDuration").toLong();
            }
  
            cout << "Normalizing " << featureFileName << endl;
            FeatureFileWriter w(featureFileName, config);                             // build a featurefile writer to output the features (real features)
                
         
            // current mean and cov vectors  
            DoubleVector cMean(fr.getVectSize(),fr.getVectSize()),cCov(fr.getVectSize(),fr.getVectSize());             
            frameCount = 0;
            // Loop on each feature
            
            // AT START ZERO PADDING + COPY OF LOOKHEAD DATA
            if(initWithDelay)
            {
                while(frameCount < windowDuration)
                {   
                    Feature &f = *new Feature(fr.getVectSize());
                    
                    //ZERO PADDING
                    if ( frameCount < (windowDuration - lookHead))
                    {
                        f.reset(); 
                    }
                    else 
                    {
                        if(!fr.readFeature(f))
                        {       
                                //Case when lookhead > file size : window size = file size
                                cout << "Reach end of file, new window size :  " << frameCount<<endl;
                                windowDuration=frameCount;
                                break;
                        }
                    }
                
                    windowFrameAcc.addObject(f);
                    frameCount++;
                }
                
                
                loadMeanAndCovParam(windowFrameAcc,cMean,cCov,windowDuration);
                //END FILL 1st WINDOW (buffer)
                  
            }
            // AT START INIT WITH A MEAN VECTOR
            else
            {
                 // TO DO  
            }
            //START NORM        
            fr.seekFeature(0);
            frameCount = 0;
           
            Feature ftmp;
            //NORMALISE FEATURES                    
            while(fr.readFeature(ftmp))
            {    
                Feature &f = *new Feature(ftmp);
                frameCount++;
                //UPDATE PARAMETERS 
                updateMeanAndCovParam(f, cMean,cCov, windowDuration, frameCount, lookHead );
                //computeCMVparameters(f, windowDuration, windowFrameAcc, cMean, cCov, frameCount  );
                //NORMALISE FEATURE                 
                computeCMVnorm(f, cMean, cCov);
                //cout <<"Frame Count :"<<frameCount<<endl;
                w.writeFeature(f);
                delete &f; //DELETE THIS IF USE OF computeCMVparameters
                
            }
            windowFrameAcc.deleteAllObjects();
            
            
        }//END LOOP ON FILES
 }      
                

  catch (Exception & e)
  {
      cout << e.toString ().c_str () << endl;
  }
  return 0;
}
                
                

#endif // !defined(ALIZE_NormFeatWindowMode_cpp)