MixtureGFStat.cpp

Go to the documentation of this file.

/*
        This file is part of ALIZE which is an open-source tool for 
        speaker recognition.

    ALIZE 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.

    ALIZE 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 ALIZE.
    If not, see <http://www.gnu.org/licenses/>.
        
        ALIZE is a development project initiated by the ELISA consortium
        [alize.univ-avignon.fr/] and funded by the French Research 
        Ministry in the framework of the TECHNOLANGUE program 
        [www.technolangue.net]

        The ALIZE 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 person=92s voice or to identify with absolute 
        certainty an individual from his or her voice.]
        Contact Jean-Francois Bonastre for more information about the licence or
        the use of ALIZE

        Copyright (C) 2003-2010
        Laboratoire d'informatique d'Avignon [lia.univ-avignon.fr]
        ALIZE admin [alize@univ-avignon.fr]
        Jean-Francois Bonastre [jean-francois.bonastre@univ-avignon.fr]
*/

#if !defined(ALIZE_MixtureGFStat_cpp)
#define ALIZE_MixtureGFStat_cpp

#include <new>
#include "MixtureGFStat.h"
#include "alizeString.h"
#include "Feature.h"
#include "DistribGF.h"
#include "Mixture.h"
#include "MixtureGF.h"
#include "DistribRefVector.h"
#include "Config.h"
#include "Exception.h"
#include "StatServer.h"

using namespace alize;
typedef MixtureGFStat M;

//-------------------------------------------------------------------------
M::MixtureGFStat(const K&, StatServer& ss, const MixtureGF& m, const Config& c)
:MixtureStat(ss, m, c), _pMixForAccumulation(NULL), _pMixtureForEM(NULL) {}
//-------------------------------------------------------------------------
MixtureGFStat& M::create(const K&, StatServer& ss,
                                     const MixtureGF& m, const Config& c)
{
  MixtureGFStat* p = new (std::nothrow) MixtureGFStat(K::k, ss, m, c);
  assertMemoryIsAllocated(p, __FILE__, __LINE__);
  return *p;
}
//-------------------------------------------------------------------------
void M::resetEM()
{
  assert(_pMixture->getDistribCount() == _distribCount);
  resetOcc();

  // copy the original mixture and its ditributions
  if (_pMixtureForEM != NULL)
    delete _pMixtureForEM;
  _pMixtureForEM = &static_cast<MixtureGF&>(_pMixture->duplicate(K::k,
                                            DUPL_DISTRIB));
  // create and initialize a temporary mixtureGF to accumulate cov and mean
  if (_pMixForAccumulation != NULL)
    delete _pMixForAccumulation;
  _pMixForAccumulation = &MixtureGF::create(K::k, "",
                      _pMixtureForEM->getVectSize(),
                      _pMixtureForEM->getDistribCount());
  unsigned long vectSize = _pMixture->getVectSize();
  for (unsigned long cc=0; cc<_distribCount; cc++)
  {
    DistribGF& d = _pMixForAccumulation->getDistrib(cc);

    real_t* m = d.getMeanVect().getArray();     
    real_t* c = d.getCovMatrix().getArray();
        
    for (unsigned long i=0; i<vectSize; i++)
    {
      m[i] = 0.0;
      for (unsigned long j=0; j<vectSize; j++)
        c[i + j*vectSize] = 0.0;
        c[0 + 0*vectSize] = 1e200; // to avoid throwing exception "Matrix is not positive definite"
    }
        
  }
  _featureCounterForEM = 0.0;
  _resetedEM = true;
}

//-------------------------------------------------------------------------
occ_t M::computeAndAccumulateEM(const Feature& f, double w)
{
  assertResetEMDone();
  real_t sum = computeAndAccumulateOcc(f, w);
  Feature::data_t* dataVect = f.getDataVector();
  unsigned long vectSize = _pMixture->getVectSize();
  unsigned long vectSize2 = vectSize*vectSize;

  for (unsigned long c=0; c<_distribCount; c++)
  {
    DistribGF& d = _pMixForAccumulation->getDistrib(c);
    real_t* dTmpMeanVect = d.getMeanVect().getArray();
    real_t* dTmpCovMatr  = d.getCovMatrix().getArray();
    
    for (unsigned long i=0; i<vectSize; i++)
    {
      real_t mean = _occVect[c] * dataVect[i];
      dTmpMeanVect[i] += mean;
      for (unsigned long j=i*vectSize; j<vectSize2; j += vectSize)
        dTmpCovMatr[i+j]  += mean * dataVect[j];
    }
  }
    _featureCounterForEM += w;
  return sum;
}
//-------------------------------------------------------------------------
void M::addAccEM(const MixtureStat& mx)
{
  const MixtureGFStat* p = dynamic_cast<const MixtureGFStat*>(&mx);
  if (p == NULL)
    throw Exception("MixtureStat incompatibility", __FILE__, __LINE__);
  if (p->_distribCount != _distribCount)
    throw Exception("MixtureStat incompatibility", __FILE__, __LINE__);
  //const MixtureGFStat& m = static_cast<const MixtureGFStat&>(mx);

  throw Exception("unimplemented method :o(", __FILE__, __LINE__);
  // TODO : implement this method
}
//-------------------------------------------------------------------------
const Mixture& M::getEM()
{
  assertResetEMDone();
  unsigned long vectSize = _pMixture->getVectSize();
  unsigned long c, idx, vectSize2 = vectSize*vectSize;
  occ_t occ, totOcc = 0.0;
  real_t* dTmpCovMatr;
  real_t* dTmpMeanVect;
  real_t* dCovMatr;
  real_t* dMeanVect;
  real_t mean, mean2, cov;

  for (c=0; c<_distribCount; c++)
    totOcc += _accumulatedOccVect[c];

  for (c=0; c<_distribCount; c++)
  {
    occ = _accumulatedOccVect[c];
    if (occ > 0.0)
    {
      DistribGF& dTmp = _pMixForAccumulation->getDistrib(c);
      dTmpCovMatr  = dTmp.getCovMatrix().getArray();
      dTmpMeanVect = dTmp.getMeanVect().getArray();

      DistribGF& d = _pMixtureForEM->getDistrib(c);
      dCovMatr  = d.getCovMatrix().getArray();
      dMeanVect = d.getMeanVect().getArray();


      for (unsigned long i=0; i<vectSize; i++)
      {
        dMeanVect[i] = mean = dTmpMeanVect[i] / occ;
        mean2 = mean*mean;
        for (unsigned long j=0; j<vectSize2; j += vectSize)
        {
          idx = i+j;
          cov  = dTmpCovMatr[idx] / occ - mean2;
          if (cov >= MIN_COV )
            dCovMatr[idx] = cov;
          else
            dCovMatr[idx] = MIN_COV;
        }
      }
      _pMixtureForEM->weight(c) = occ/totOcc;
      d.computeAll();
    }
  }
  return *_pMixtureForEM;
}
//-------------------------------------------------------------------------
MixtureGF& M::getInternalAccumEM()
{
  assertResetEMDone();
  assert(_pMixForAccumulation != NULL);
  return *_pMixForAccumulation;
}
//-------------------------------------------------------------------------
String M::getClassName() const { return "MixtureGFStat"; }
//-------------------------------------------------------------------------
M::~MixtureGFStat()
{
  if (_pMixForAccumulation != NULL)
    delete _pMixForAccumulation;
  if (_pMixtureForEM != NULL)
    delete _pMixtureForEM;
}
//-------------------------------------------------------------------------

#endif // !defined(ALIZE_MixtureGFStat_cpp)