ScoreWarp.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_ScoreWarp_cpp)
#define ALIZE_ScoreWarp_cpp
 
#include <iostream>
#include <fstream>
#include <cstdio>
#include <cassert>
#include <cmath>
#include "TrainTools.h"
#include "ScoreWarp.h"
const double PI2 = 3.14159265358979323846*2;
static double x1;
static double x2;
void boxMullerGeneratorInit(){
 x1= (rand()/ (float)RAND_MAX);
}
double boxMullerGenerator(double mean, double cov){
  x2=x1;
  x1= (rand()/ (float)RAND_MAX);
  if (debug) cout <<"boxMullerGenerator x1["<<x1<<"] x2["<<x2<<"]"<<endl;
  double y= sqrt(-2.0*log(x1))*cos(PI2*x2);   // Box-Muller gaussian generator from 2 sample of [0,1] indep. random numbers
  double ret= (y*cov)+mean;                        // fit to the N(mean,cov) distribution
  if (debug) cout <<"boxMullerGenerator["<<ret<<"]"<<endl;
  return ret;
}


// Build the Gaussian target distribution (mean,cov), by generating nbSample data, specifying the number of bins
Histo makeGausHisto(unsigned long nbSample,double mean, double cov,unsigned long nbBins){
  if (verbose) cout << "makeGaussHisto, nbSample["<<nbSample<<"] mean["<<mean<<"] cov["<<cov<<"] nbBin["<<nbBins<<"]"<<endl;
  Histo histo(nbBins);
  boxMullerGeneratorInit();       // Init The box-muller gaussian number generator
  for (unsigned long i=0;i<nbSample;i++)
    histo.accumulateValue(boxMullerGenerator(mean,cov));
  histo.computeHisto(); 
  if (verbose){
    double tot=0;
    for (unsigned long i=0;i<histo.size();i++) 
      tot+=histo.count(i)*(histo.higherBound(i)-histo.lowerBound(i));
    cout <<"makeGaussHisto tot of the final pdf["<<tot<<"]"<<endl;
  }
  return histo;
}


double centralSpace(const Histo &warpH,double a){
  if (a==0) return 0;
  unsigned long inf,sup;
  double t=(1.0-a)/2.0;
  inf=0;
  for (double tot=0.0;(inf<warpH.size()) && (tot<t);inf++)
    tot+=areaHisto(warpH,inf);
  sup=warpH.size()-1;
  for (double tot=1.0;(sup>=0) && (tot>1-t);sup--)
    tot-=areaHisto(warpH,sup);
  return (warpH.lowerBound(sup)-warpH.higherBound(inf));
}

// Compute the warped score, using the raw score distribution warH and the destination distribution destH
double scoreWarping(double score, const Histo& warpH, const Histo& destH, double nonObserved,double refArea)
{
  if (debug) cout << "scoreWarping score ["<<score<<"] nonObserved["<<nonObserved<<"] refArea["<<refArea<<"] ";
  // Value before the min, after the max
  if (score<warpH.lowerBound(0)){                // the value is less than the minimum of the raw score distribution 
    double infBound=warpH.lowerBound(0)-centralSpace(warpH,refArea);
    return destH.lowerBound(0)-(linearInterpolation(score,infBound,warpH.lowerBound(0))*centralSpace(destH,refArea));
  } 
  if (score>warpH.higherBound(warpH.size()-1)){  // the value is more than the maximum of the raw distribution
    double supBound=warpH.higherBound(warpH.size()-1)+centralSpace(warpH,refArea);
    return destH.higherBound(destH.size()-1)+(linearInterpolation(score,warpH.higherBound(warpH.size()-1),supBound)*centralSpace(destH,refArea)); 
  }
  
  double totalWarp=0.0;
  // Compute the area between -infinite and the score (raw distrib) - totalWarp
  unsigned long idxW=0;                          // Will be the bin number where is the score (in the raw distrib)
  for(;(idxW<warpH.size())&&(warpH.higherBound(idxW)<score); idxW++);
  if (debug) cout << " idxW["<<idxW<<"]";
  totalWarp=nonObserved/2;                       // nonObserved is the estimated amount of data non seen in the set
  for (unsigned long idx=0;idx<idxW;idx++)       // Accumulate the area for the raw distrib (numerical integration)
    totalWarp+=areaHisto(warpH,idx);             // Without the last bin
  double percentW=linearInterpolation(score,warpH.lowerBound(idxW),warpH.higherBound(idxW));
  if (debug) cout << " percentW["<<percentW<<"]";
  totalWarp+=areaHisto(warpH,idxW)*percentW;    // Add a percentage of the last bin (linear interpolation)
  if (debug) cout << " totalW["<<totalWarp<<"]";

  // Find idxD, the index of the corresponding bin (area bin[-infinite] until bin[idxD]=totalWarp
  double totalDest;
  unsigned long idxD; 
  for (idxD=0,totalDest=0;(idxD<destH.size())&&(totalDest<totalWarp);idxD++)
    totalDest+=areaHisto(destH,idxD);
  if (idxD==destH.size()){
    idxD--;
    totalDest-=areaHisto(destH,idxD);
  }
  else
    if (idxD>0){
      totalDest-=areaHisto(destH,idxD);
      idxD--;
    }    
  // Compute the final score
  double ret=destH.lowerBound(idxD);                                                       // Set the ret to the lowerbound of the good bin
  double percentH=linearInterpolation(totalWarp,totalDest,areaHisto(destH,idxD)+totalDest);// Compute the linear interpolation % for the last bin
  if (debug) cout << " idxD["<<idxD<<"] TotalDet["<<totalDest<<"] PercentD["<<percentH<<"]";
  ret+=(destH.higherBound(idxD)-destH.lowerBound(idxD))*percentH;                          // apply the interpolation
  if (debug) cout <<" final["<<ret<<"]"<<endl;
  return ret;
}
 

// Compute the warped value using score, data histo warpH,
// target histo destH
// it assumes that all the values are in warpH bounds

double warping(double score, const Histo& warpH, const Histo& destH)
{
  if (debug) cout << "warping input ["<<score<<"] ";
  
  double totalWarp=0.0;
  // Compute the area between -infinite and the score (raw distrib) - totalWarp
  unsigned long idxW=0;                          // Will be the bin number where is the score (in the raw distrib)
  for(;(idxW<warpH.size())&&(warpH.higherBound(idxW)<score); idxW++);
  if (debug) cout << " idxW["<<idxW<<"] ";

  for (unsigned long idx=0;idx<idxW;idx++)       // Accumulate the area for the raw distrib (numerical integration)
    totalWarp+=areaHisto(warpH,idx);             // Without the last bin
  double percentW=linearInterpolation(score,warpH.lowerBound(idxW),warpH.higherBound(idxW));
  if (debug) cout << " percentW["<<percentW<<"] ";
  totalWarp+=areaHisto(warpH,idxW)*percentW;    // Add a percentage of the last bin (linear interpolation)
  if (debug) cout << " totalW["<<totalWarp<<"] ";

  // Find idxD, the index of the corresponding bin (area bin[-infinite] until bin[idxD]=totalWarp
  double totalDest;
  unsigned long idxD; 
  for (idxD=0,totalDest=0;(idxD<destH.size())&&(totalDest<totalWarp);idxD++)
    totalDest+=areaHisto(destH,idxD);
  if (idxD==destH.size()){
    idxD--;
    totalDest-=areaHisto(destH,idxD);
  }
  else
    if (idxD>0){
      totalDest-=areaHisto(destH,idxD);
      idxD--;
    }    
  // Compute the final value
  double ret=destH.lowerBound(idxD);                                                       // Set the ret to the lowerbound of the good bin
  double percentH=linearInterpolation(totalWarp,totalDest,areaHisto(destH,idxD)+totalDest);// Compute the linear interpolation % for the last bin
  if (debug) cout << " idxD["<<idxD<<"] TotalDet["<<totalDest<<"] Percent last bin["<<percentH<<"] ";
  ret+=(destH.higherBound(idxD)-destH.lowerBound(idxD))*percentH;                          // apply the interpolation
  if (debug) cout <<" warping output["<<ret<<"] "<<endl;
  return ret;
}

#endif // !defined(ALIZE_ScoreWarp_cpp)