UnsupervisedTools.h

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_UnsupervisedTools_h)
#define ALIZE_UnsupervisedTools_h

#if defined(_WIN32)
#if defined(LIA_SPKTOOLS_EXPORTS)
#define LIA_SPKTOOLS_API __declspec(dllexport)
#else
#define LIA_SPKTOOLS_API __declspec(dllimport)
#endif
#else
#define LIA_SPKTOOLS_API
#endif

#include <alize.h>
#include "liatools.h"

using namespace alize;
using namespace std;



//Accumulate statistics on a frame, a segment,a cluster.
LIA_SPKTOOLS_API void accumulateStatLK(StatServer & ss, FeatureServer & fs, MixtureStat & acc, unsigned long idxBeginFrame, unsigned long nbFrames, Config & config);
LIA_SPKTOOLS_API void accumulateStatLK(StatServer & ss, FeatureServer & fs, MixtureStat & acc, Seg * seg, Config & config);
LIA_SPKTOOLS_API void accumulateStatLK(StatServer & ss, FeatureServer & fs, MixtureStat & acc, SegCluster & selectedSegments, Config & config);




// The main function for estimate a client model by bayesian adaptattion of a world model
// Using EM and MAP 
//void adaptModel(Config& config,StatServer &ss,MixtureServer &ms,FeatureServer &fs,SegCluster& selectedSegments,MixtureGD &world,MixtureGD &clientMixture);//A.P.
LIA_SPKTOOLS_API void adaptModelEM(Config& config,StatServer &ss,MixtureServer &ms,FeatureServer &fs,SegCluster& selectedSegments,MixtureGD &world,MixtureGD &clientMixture);//A.P.
LIA_SPKTOOLS_API void adaptModelEM(Config& config,StatServer &ss,MixtureServer &ms,FeatureServer &fs,SegCluster& selectedSegments,FeatureServer &fs2,SegCluster& selectedSegments2,MixtureGD &aprioriModel,MixtureGD &clientMixture,MAPCfg &mapCfg);
LIA_SPKTOOLS_API void adaptModelEM(Config& config,StatServer &ss,MixtureServer &ms,FeatureServer &fs,SegCluster& selectedSegments,FeatureServer &fs2,SegCluster& selectedSegments2,
                MixtureGD &aprioriModel,MixtureGD &clientMixture);
LIA_SPKTOOLS_API void adaptModelMAP(Config& config,StatServer &ss,MixtureServer &ms,MixtureGD &world,MixtureGD &clientMixture, unsigned long &frameCount);
LIA_SPKTOOLS_API void adaptModelMAP(Config & config, StatServer & ss, MixtureServer & ms, MixtureGD & aprioriModel, MixtureGD & clientMixture, MAPCfg & mapCfg,  unsigned long &frameCount);
//void adaptModel(Config& config,StatServer &ss,MixtureServer &ms,FeatureServer &fs,SegCluster& selectedSegments,MixtureGD &world,MixtureGD &clientMixture, MAPCfg &cfg);
// ** New training algo based on a true EM/ML estimate of the training data before to apply MAP
LIA_SPKTOOLS_API void modelBasedadaptModel(Config& config,StatServer &ss,MixtureServer &ms,FeatureServer &fs,SegCluster& selectedSegments,
                          MixtureGD &aprioriModel,MixtureGD &clientMixture, MixtureGD &initModel);
LIA_SPKTOOLS_API void modelBasedadaptModelEM(Config& config,StatServer &ss,MixtureServer &ms,FeatureServer &fs,SegCluster& selectedSegments,
                          MixtureGD &aprioriModel,MixtureGD &clientMixture, MixtureGD &initModel);
LIA_SPKTOOLS_API void modelBasedadaptModelEM(Config& config,StatServer &ss,MixtureServer &ms,FeatureServer &fs,SegCluster& selectedSegments,FeatureServer &fsTests,SegCluster& selectedSegmentsTests,MixtureGD &aprioriModel,MixtureGD &clientMixture, MixtureGD &initModel);
LIA_SPKTOOLS_API void adaptModel(Config& config,StatServer &ss,MixtureServer &ms,ObjectRefVector &FeatServ,ObjectRefVector & ClusterSeg,
                MixtureGD &aprioriModel,MixtureGD &clientMixture,DoubleVector &decision);




//Compute LLR 
//basic
LIA_SPKTOOLS_API double computeLLR(StatServer & ss, FeatureServer & fsTests, MixtureGD & world,
                MixtureGD & clientMixture, SegCluster & selectedSegmentsTests);
//basic and write top information 
LIA_SPKTOOLS_API double computeLLR(Config & config, StatServer & ss, FeatureServer & fsTests,  MixtureGD & world, MixtureGD & clientMixture,
                SegCluster & selectedSegmentsTests, String & idTest);
//Fast by using top information
LIA_SPKTOOLS_API double computeFastLLR(StatServer & ss, FeatureServer & fsTests, MixtureGD & world, MixtureGD & clientMixture, SegCluster & selectedSegmentsTests,
                String & idTest, Config & config);
//modelbased
LIA_SPKTOOLS_API double computeLLRGD(Config & config, MixtureGD & clientMixture, MixtureGD & world,  MixtureGD & dataTest);


//compute weights used for adaptation for each tests
//logistic regression, take parameter from the config
LIA_SPKTOOLS_API void expandLLR(DoubleVector & decision, Config & configTest);
//WMAP using gaussian formulae
LIA_SPKTOOLS_API void WMAP(DoubleVector & decision, Config & configTest);
//WMAP on GMM on scores
LIA_SPKTOOLS_API void WMAPGMM(DoubleVector & decision, Config & configTest, MixtureGD & tar,  MixtureGD & non, StatServer & ss);
LIA_SPKTOOLS_API void WMAPGMMFixedPriors(DoubleVector & decision, Config & configTest, MixtureGD & tar,
                MixtureGD & non, StatServer & ss);
LIA_SPKTOOLS_API String getFullFileName(String & id, Config & c);
LIA_SPKTOOLS_API String getFullMixtureName(String & id, Config & c);
LIA_SPKTOOLS_API bool FileExists(String & fullFileName);


//compute a MAP model from EM models (fast)
//takes a list of EM models, the number of frames each model was learnt with, and a weight for each model  
LIA_SPKTOOLS_API void computeMAPmodelFromEMones(Config & config, StatServer & ss,  MixtureServer & ms, DoubleVector & nbFramesSelected,
                MixtureGD & aprioriModel, MixtureGD & clientMixture, MixtureGD & aux,  MixtureGD & tmp, DoubleVector & decision, XLine & testsToCompute);

LIA_SPKTOOLS_API double SegClusterFrame(SegCluster & SegC); //return the number of frames in a cluster

LIA_SPKTOOLS_API void addLineInXList(XLine &line,XList &list);  

LIA_SPKTOOLS_API void computePriors(DoubleVector & decision, DoubleVector &priorImp, DoubleVector &priorTar,Config & configTest); //update Imp and Tar priors during the adaptation process


//load TNORM or ZNORM parameters, need trials scores on impostor models, a list of the trials name.

//load impostors scores for TNORM computation from imp_seg.res file
//Must concatenate imp_seg.res and imp_imp.res if you want to do ZTNORM.
LIA_SPKTOOLS_API void loadTnormParam(String &inputClientListFileName,String &testFile,ObjectRefVector &stockTnorm,Config &config);



//Compute Znorm "online", !!!!!could do the same for TNORM but not implemented yet!!!!!
LIA_SPKTOOLS_API void computeAndStoreZnormParam(StatServer &ss, String & inputImpListFileName, String &idclient, MixtureGD &clientMixture,
                ObjectRefVector & stockZnorm, MixtureGD &world, Config & config, bool &ztnorm, ObjectRefVector & stockTnorm);

//To create Impostor models once
LIA_SPKTOOLS_API void learnMAPimpostorModels(Config & config, DoubleVector &NbFramesSelectedImp,MixtureServer &ms,StatServer &ss,MixtureGD &world);  


//Normalize a score, TNORM or ZNORM : the same function is used
LIA_SPKTOOLS_API void normalizeScore(String &test, double &decision, ObjectRefVector &stockNorm);
LIA_SPKTOOLS_API void normalizeScore(String &test, double &decision, ObjectRefVector &stockNorm, double &shift);

//Oracle TEST 
LIA_SPKTOOLS_API void resetWeights(DoubleVector & decision);
LIA_SPKTOOLS_API void Oracle(String &idTar, String & idTest, double &score, Config & config,MixtureGD & tar,
                MixtureGD & non, StatServer & ss);

//Compute a train model with a selected percentage of train data, and compute the LLR with the rest of the data (this is done several times ) return the EM estimate and the SegCluster associated
//for the smaller LLR.
LIA_SPKTOOLS_API void crossValid(Config & configTest, StatServer & ss,  MixtureServer & ms, FeatureServer & fs, SegCluster & totalSegments,
                MixtureGD & aprioriModel,MixtureGD & bestModel,SegCluster & selectedSegments,String & idTest);

//2 purposes : do not recalculate LLR of test data on a client model, and possibility to use RES file from a different recognition system to compute adaptation weights
LIA_SPKTOOLS_API double searchLLRFromResFile(String &idTar, String &test ,String &inputResFilename, Config & config);

//Create an EM/ML model (1 iter) by weighting each seg thanks to WMAP
LIA_SPKTOOLS_API unsigned long adaptModelEMweightedFrames(String &labelSelectedFrames,XLine & featureFileName,StatServer &ss,MixtureGD &world, MixtureGD &tar, MixtureGD &non,MixtureGD &MixtureforLLR,MixtureGD &MixtureEMOutput, Config &config, FeatureServer &fs, String &fullFileName,String &idTest);

//Return the String id of the model in a list TARLIstFilename for which the LLR of selectedSegments is MAX. 
LIA_SPKTOOLS_API String selectNearestTarModel(String &TARListFilename, String & fullFileName,Config &config, StatServer & ss, FeatureServer & fs,
                MixtureGD & world,  SegCluster & selectedSegments,MixtureServer &ms);

//Input : ndx to create target model
//Output : matrix of crossing LLR 
LIA_SPKTOOLS_API void computeLLRmatrix(DoubleMatrix &LLR, XLine &models, XList &features, Config &config, StatServer &ss, MixtureGD & world, MixtureServer &ms,String &labelSelectedFrames);


//Method for weighted fusion of MAP models 
LIA_SPKTOOLS_API void fuseMAP(Config & config, StatServer & ss,  MixtureServer & ms, MixtureGD & aprioriModel, MixtureGD & clientMixture, MixtureGD & aux,
                MixtureGD & tmp, DoubleVector & decision, XLine & testsToCompute);
//weighted sum of MAP means of two GMM
LIA_SPKTOOLS_API void fuseMAPMeans(const MixtureGD & model1,double &w1,const MixtureGD & model2,double &w2,MixtureGD &result);


LIA_SPKTOOLS_API double computeWeightedLLR(Config & config, StatServer & ss, FeatureServer &fsTests, MixtureServer & ms, MixtureGD & aprioriModel, SegCluster & selectedSegmentsTests,MixtureGD & aux, DoubleVector & decision, XLine & testsToCompute, String &idTest, String &fullfilename);

//to copy cluster
LIA_SPKTOOLS_API void copyCluster(SegCluster &source, SegCluster &dest);
LIA_SPKTOOLS_API int findClusterInRefvector(ObjectRefVector &SegServ,String & id);


//ADAPTATIVE TNORM functions

//compute all EM ML models for the cohort and store it (one session training)
LIA_SPKTOOLS_API void learnEMimpostorModels(Config & config, DoubleVector &NbFramesSelectedImp,MixtureServer &ms,StatServer &ss,MixtureGD &world);
//load Tnorms models, update it and compute Tnorm on a score
LIA_SPKTOOLS_API double computeAdaptedTnorm(Config & configTest, MixtureServer & ms,StatServer &ss, MixtureGD &world, MixtureGD &auxMixture, MixtureGD &tmpMixture, double &nonorm_score, int &countTests, DoubleVector &decision, DoubleVector &NbFramesSelected, XLine &testsToCompute,
                DoubleVector &NbFramesSelectedImp, FeatureServer &fsTests, SegCluster &selectedSegmentsTests, String &idTest);
        
//need a matrix columns : nb session for training , rows : clients LLR
//look for the nearest LLR in the matrix (for a given nb of training session) and returns the delta LLR between one session and the given nb of session
LIA_SPKTOOLS_API double findNearestLLRInMatrix(Matrix <double> &Mat, unsigned long &index, double &LLR);        

//USE TRAIN DATA TO ASSESS THE ADAPATION WEIGHT
LIA_SPKTOOLS_API void assessAdaptation(StatServer &ss,MixtureGD & adaptedMixture ,SegCluster &selectedSegments, 
                Config &configTest, FeatureServer &fs, MixtureGD &world, DoubleVector &tmp, String &fullFileNameTrain, 
                int countTests, double &tarScore);
        

#endif //!defined(ALIZE_UnsupervisedTools_h)