TopGauss.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_TopGauss_cpp)
#define ALIZE_TopGauss_cpp

#define myTINY 0.0000001

#include <iostream>
#include <fstream>  // pour outFile
#include <cstdio>   // pour printf()
#include <cassert> // pour le debug pratique
#include <cmath>
#include "liatools.h"
#include "TopGauss.h"
#define EPS_LK 1e-200
unsigned long TopGauss::frameToIdx(unsigned long & f) {
        unsigned long cnt=0;
        unsigned long*_pnbg=_nbg.getArray();
        for (unsigned long t=0;t<f;t++)
                cnt+=_pnbg[t];
return cnt;
}

void TopGauss::read(String & featureFilename,Config &config) {
        if (verbose) cout << "(TopGauss::read) File ["<<featureFilename<<"] ... ";
        String nbGaussPerFrameFilename=config.getParam("nbGaussianFilesDir")+featureFilename;
        if (verboseLevel > 2) cout << "(TopGauss) Reading indexes in file ["<<nbGaussPerFrameFilename<<"]"<<endl;       
        ifstream nbGaussFile(nbGaussPerFrameFilename.c_str(),ios::in|ios::binary);                              //ios::binary added by alarcher
        if(!nbGaussFile){throw Exception("Cannot find nbGaussian file", __FILE__, __LINE__);}   
        // Read number of frames
        nbGaussFile.read((char *)&_nt,sizeof(unsigned long));
        nbGaussFile.read((char *)&_nbgcnt,sizeof(unsigned long));       
        _idx.setAllValues(0);_idx.setSize(_nbgcnt);
        _nbg.setAllValues(0);_nbg.setSize(_nt);
        _snsl.setAllValues(0);_snsl.setSize(_nt);
        _snsw.setAllValues(0);_snsw.setSize(_nt);       
        nbGaussFile.read((char *)_nbg.getArray(),sizeof(unsigned long)*_nt);
        nbGaussFile.read((char *)_idx.getArray(),sizeof(unsigned long)*_nbgcnt);
        nbGaussFile.read((char *)_snsw.getArray(),sizeof(double)*_nt);
        nbGaussFile.read((char *)_snsl.getArray(),sizeof(double)*_nt);  
        nbGaussFile.close();
}

void TopGauss::write(String & featureFilename,Config & config) {
        if (verbose) cout << "(TopGauss::write) File ["<<featureFilename<<"] "<<endl;   
        String nbGaussFilename=config.getParam("nbGaussianFilesDir")+featureFilename;
        ofstream nbGaussFile (nbGaussFilename.c_str(),ios::out|ios::binary);                    //ios::binary added by alarcher
        if(!nbGaussFile){throw Exception("Cannot find nbGaussian file", __FILE__, __LINE__);}   
        nbGaussFile.write((char *)&_nt,sizeof(unsigned long));          // write number of frames in cluster
        nbGaussFile.write((char *)&_nbgcnt,sizeof(unsigned long));              // write number of frames in cluster    
        // Write number of Gaussians + Gaussian Indexes;
        nbGaussFile.write((char *)_nbg.getArray(),sizeof(unsigned long)*_nbg.size());
        nbGaussFile.write((char *)_idx.getArray(),sizeof(unsigned long)*_idx.size());
        nbGaussFile.write((char *)_snsw.getArray(),sizeof(double)*_snsw.size());
        nbGaussFile.write((char *)_snsl.getArray(),sizeof(double)*_snsl.size());        
        nbGaussFile.close();
}

// Init with a Filename 
double TopGauss::compute(MixtureGD & UBM,String & featureFilename, Config & config) {
        if (verbose) cout << "(TopGauss::compute on a single file) File ["<<featureFilename<<"] ... ";
        FeatureServer fs(config,featureFilename);  
        double llk=compute(UBM,fs,featureFilename,config);      
        if (verbose) cout <<_nbgcnt/_nt <<" per frames LLK="<<llk<<endl;        
return llk;
}

//Init with a FeatureServer
RealVector <double> TopGauss::compute(MixtureGD & UBM,FeatureServer &fs,Config & config) {
        RealVector <double> llkv;
        for (unsigned long i=0;i<fs.getSourceCount();i++) {
                String filename=fs.getNameOfASource(i);
                if (verbose) cout << "(TopGauss::compute on FeatureServer) File ["<<filename<<"] ... ";                 
                llkv.addValue(compute(UBM,fs,filename,config));
                cout << "_nt"<<_nt<<" idx"<<_idx.size()<<" _snsw"<<_snsw.size()<<endl;
                this->write(filename,config); // this should be removed!! but when passing a feature server things get worse            
                if (verbose) cout <<_nbgcnt/_nt <<" per frames LLK="<<llkv[i]<<endl;    
        }
return llkv;
}


// Main init function
double TopGauss::compute(MixtureGD & UBM,FeatureServer &fs,String & featureFilename,Config & config){
        StatServer ss(config);
        MixtureGDStat &acc=ss.createAndStoreMixtureStat(UBM);   
        unsigned long _mixsize=UBM.getDistribCount();
        String labelSelectedFrames =config.getParam("labelSelectedFrames");
        unsigned long begin=fs.getFirstFeatureIndexOfASource(featureFilename);
        fs.seekFeature(begin);
        SegServer segmentsServer;
        LabelServer labelServer;
        initializeClusters(featureFilename,segmentsServer,labelServer,config);
        verifyClusterFile(segmentsServer,fs,config);
        unsigned long codeSelectedFrame=labelServer.getLabelIndexByString(labelSelectedFrames); 
        SegCluster& selectedSegments=segmentsServer.getCluster(codeSelectedFrame);  
        acc.resetLLK();
        double topD=config.getParam("topGauss").toDouble();
        if (verbose) {if(topD<1.0) cout << "LLK %="<< topD << "% ";else cout << "Top-"<<topD<<" ";}
        
        // Class values
        _nt=totalFrame(selectedSegments);       
        _nbg.setSize(_nt); _idx.setSize(0);_snsw.setSize(0); _snsl.setSize(0);
        _nbg.setAllValues(0); _idx.setAllValues(0);_snsw.setAllValues(0.0);_snsl.setAllValues(0.0);
        _nbgcnt=0;
        Seg *seg;          // current selected segment
        selectedSegments.rewind();              
        unsigned long t=0; //cnt frames
        while((seg=selectedSegments.getSeg())!=NULL){                           
                unsigned long begin=seg->begin()+fs.getFirstFeatureIndexOfASource(seg->sourceName()); 
                fs.seekFeature(begin);
                Feature f;
                for (unsigned long idxFrame=0;idxFrame<seg->length();idxFrame++){
                        fs.readFeature(f); 
                        double llk=acc.computeAndAccumulateLLK(f,1.0,DETERMINE_TOP_DISTRIBS);
                        const LKVector &topV=ss.getTopDistribIndexVector();
                        double lk_tot=exp(llk);
                        
                        double val=0.0;
                        if (topD<1.0) {
                                for(unsigned long j=0;j<_mixsize;j++){
                                        if (val > topD*lk_tot) break;
                                        val+=(topV[j].lk);
                                        _nbg[t]++;
                                }
                        } else _nbg[t]=(unsigned long)topD;
                        _nbgcnt+=_nbg[t];
                         
                        double snsw=1.0;
                        double snsl=lk_tot;                                     
                        for(unsigned long j=0;j<_nbg[t];j++) {
                                _idx.addValue(topV[j].idx);             
                                snsw -=UBM.weight(topV[j].idx);
                                snsl -=topV[j].lk;
                        }

                        _snsw.addValue(snsw);
                        if (snsl < EPS_LK)
                                _snsl.addValue(EPS_LK);
                        else _snsl.addValue(snsl);
                        t++;
                }               
        }
        if (t!=_nt) cout << "W: t("<<t<<") != _nt(" <<_nt<<")"<<endl;
return acc.getMeanLLK();
}



RealVector <double> TopGauss::get(MixtureGD & UBM,FeatureServer &fs,Config & config){
        RealVector <double> llkv;
        for (unsigned long i=0;i<fs.getSourceCount();i++) {
                String filename=fs.getNameOfASource(i);
                if (verbose) cout << "(TopGauss::get on FeatureServer) File ["<<filename<<"] ... ";     
                this->read(filename,config); // this should be removed!! but when passing a feature server things get worse                     
                llkv.addValue(get(UBM,fs,filename,config));
                if (verbose) cout <<_nbgcnt/_nt <<" per frames LLK="<<llkv[i]<<endl;    
        }
return llkv;
}

// Can use this function to get likelihood with a topgauss
double TopGauss::get(MixtureGD & UBM,String & featureFilename,Config & config){
        FeatureServer fs(config,featureFilename);  
        if (verbose) cout << "(TopGauss::get on a single file) File ["<<featureFilename<<"] ... ";      
        double llk=get(UBM,fs,featureFilename,config);  
        if (verbose) cout <<_nbgcnt/_nt <<" per frames LLK="<<llk<<endl;        
return llk;     
}

// Can use this function to get likelihood with a topgauss
/*double TopGauss::get(MixtureGD & UBM,FeatureServer &fs,String & featureFilename,Config & config){
        double llkcluster=0.0;
        StatServer ss(config);
        String labelSelectedFrames =config.getParam("labelSelectedFrames");
        unsigned long begin=fs.getFirstFeatureIndexOfASource(featureFilename);
        fs.seekFeature(begin);
        SegServer segmentsServer;
        LabelServer labelServer;
        initializeClusters(featureFilename,segmentsServer,labelServer,config);
        verifyClusterFile(segmentsServer,fs,config);
        unsigned long codeSelectedFrame=labelServer.getLabelIndexByString(labelSelectedFrames); 
        SegCluster& selectedSegments=segmentsServer.getCluster(codeSelectedFrame);  
        
        Seg *seg;          // current selected segment
        selectedSegments.rewind();              
        unsigned long t=0; //cnt frames
        double minLLK=config.getParam("minLLK").toDouble();
        double maxLLK=config.getParam("maxLLK").toDouble();     
        while((seg=selectedSegments.getSeg())!=NULL){  
                double llkseg=0.0;
                unsigned long begin=seg->begin()+fs.getFirstFeatureIndexOfASource(seg->sourceName()); 
                fs.seekFeature(begin);
                Feature f;
                for (unsigned long idxFrame=0;idxFrame<seg->length();idxFrame++){
                        double lkframe=0.0;
                        fs.readFeature(f); 
                        unsigned long idx=this->frameToIdx(t);
                        unsigned long nbg=_nbg[t];
                        for (unsigned long i=0;i<nbg;i++) {
                                double lkdistrib=0.0;
                                lkdistrib=(UBM.weight(_idx[idx+i])*UBM.getDistrib(_idx[idx+i]).computeLK(f));
                                if (std::isnan(lkdistrib)) lkdistrib=exp(minLLK);
                                lkframe+=lkdistrib;
                        }
                        if (log(lkframe) <= minLLK) llkseg+=minLLK;
                        else if(log(lkframe) >= maxLLK) llkseg+=maxLLK;
                        else llkseg+=log(lkframe);                      
                        t++;
                }       
                llkcluster+=llkseg;
        }
        if (t!=_nt) cout << "W: t("<<t<<") != _nt(" <<_nt<<")"<<endl;
return llkcluster/totalFrame(selectedSegments);
}*/

// Can use this function to get likelihood with a topgauss
double TopGauss::get(MixtureGD & UBM,FeatureServer &fs,String & featureFilename,Config & config){
        StatServer ss(config);
        String labelSelectedFrames =config.getParam("labelSelectedFrames");
        unsigned long begin=fs.getFirstFeatureIndexOfASource(featureFilename);
        fs.seekFeature(begin);
        SegServer segmentsServer;
        LabelServer labelServer;
        initializeClusters(featureFilename,segmentsServer,labelServer,config);
        verifyClusterFile(segmentsServer,fs,config);
        unsigned long codeSelectedFrame=labelServer.getLabelIndexByString(labelSelectedFrames); 
        SegCluster& selectedSegments=segmentsServer.getCluster(codeSelectedFrame);  
        MixtureGDStat &acc=ss.createAndStoreMixtureStat(UBM);
        
        Seg *seg;          // current selected segment
        selectedSegments.rewind();              
        unsigned long t=0; //cnt frames
        acc.resetLLK();
        unsigned long idxBegin=0;
        while((seg=selectedSegments.getSeg())!=NULL){  
                unsigned long begin=seg->begin()+fs.getFirstFeatureIndexOfASource(seg->sourceName()); 
                fs.seekFeature(begin);
                Feature f;
                idxBegin=this->frameToIdx(t);
                for (unsigned long idxFrame=0;idxFrame<seg->length();idxFrame++){
                        fs.readFeature(f); 
                        //unsigned long idx=this->frameToIdx(t);
                        unsigned long nbg=_nbg[t];      
                        ULongVector index;
                        double sumNonSelectedWeights=_snsw[t];
                        double sumNonSelectedLLK=_snsl[t];
                        for (unsigned long i=0;i<nbg;i++) {
                                index.addValue(_idx[idxBegin+i]);
                        }               
                        char c[100];
                        sprintf(c,"%d",(int)index.size());
                        config.setParam("topDistribsCount",c); // this should be high enough    
                        if (t==0) {acc.computeAndAccumulateLLK(f,1.0,DETERMINE_TOP_DISTRIBS);acc.resetLLK();} // to remove in ALIZE, this is to init the LKvector
                        ss.setTopDistribIndexVector(index, sumNonSelectedWeights, sumNonSelectedLLK);
                        acc.computeAndAccumulateLLK(f,1.0,USE_TOP_DISTRIBS);
                        idxBegin+=nbg;
                        t++;
                }       
        }       
        //ss.deleteMixtureStat(acc);
        if (t!=_nt || idxBegin !=_nbgcnt) cout << "W: t("<<t<<") != _nt(" <<_nt<<")"<<"W: idxBegin("<<idxBegin<<") != _nbgcnt(" <<_nbgcnt<<")"<<endl;
return acc.getMeanLLK();
}

#endif