/**
* Decode.
*
* @param nbComp the nb comp
* @param threshold the threshold
* @param clusterSet the cluster set
* @param featureSet the feature set
* @param parameter the parameter
* @return the cluster set
* @throws Exception the exception
*/
public ClusterSet decode(
int nbComp,
double threshold,
ClusterSet clusterSet,
AudioFeatureSet featureSet,
Parameter parameter)
throws Exception {
String mask = parameter.getParameterSegmentationOutputFile().getMask();
String oldModelKind = parameter.getParameterModel().getModelKindAsString();
int oldNumberOfComponent = parameter.getParameterModel().getNumberOfComponents();
// ** Train GMM for each cluster.
// ** GMM is a 8 component gaussian with diagonal covariance matrix
// ** one GMM = one speaker = one cluster
// ** initialization of the GMMs :
// ** - same global covariance for each gaussian,
// ** - 1/8 for the weight,
// ** - means are initialized with the mean of 10 successive vectors taken
parameter.getParameterModel().setModelKind("DIAG");
parameter.getParameterModel().setNumberOfComponents(nbComp);
GMMArrayList gmmInitVect = new GMMArrayList(clusterSet.clusterGetSize());
MTrainInit.make(featureSet, clusterSet, gmmInitVect, parameter);
// ** EM training of the initialized GMM
GMMArrayList gmmVect = new GMMArrayList(clusterSet.clusterGetSize());
MTrainEM.make(featureSet, clusterSet, gmmInitVect, gmmVect, parameter);
// ** set the penalty to move from the state i to the state j, penalty to move from i to i is
// equal to 0
parameter.getParameterDecoder().setDecoderPenalty(String.valueOf(threshold));
// ** make Viterbi decoding using the 8-GMM set
// ** one state = one GMM = one speaker = one cluster
ClusterSet clustersDClust = MDecode.make(featureSet, clusterSet, gmmVect, parameter);
if (parameter.getParameterDiarization().isSaveAllStep()) {
parameter.getParameterSegmentationOutputFile().setMask(mask + ".d.seg");
MainTools.writeClusterSet(parameter, clustersDClust, false);
}
// ** move the boundaries of the segment in low energy part of the signal
ClusterSet clustersAdjClust = SAdjSeg.make(featureSet, clustersDClust, parameter);
if (parameter.getParameterDiarization().isSaveAllStep()) {
parameter.getParameterSegmentationOutputFile().setMask(mask + ".adj.seg");
MainTools.writeClusterSet(parameter, clustersAdjClust, false);
}
parameter.getParameterSegmentationOutputFile().setMask(mask);
parameter.getParameterModel().setNumberOfComponents(oldNumberOfComponent);
parameter.getParameterModel().setModelKind(oldModelKind);
return clustersAdjClust;
}
/**
* Tun ester2 speaker clr clustering.
*
* @param referenceClusterSet the reference cluster set
* @param uemClusterSet the uem cluster set
* @param partialKey the partial key
* @param method the method
* @param clusterSetBase the cluster set base
* @param clusterSet the cluster set
* @param featureSet the feature set
* @param parameter the parameter
* @return the diarization result list
* @throws Exception the exception
*/
public DiarizationResultList tunEster2SpeakerCLRClustering(
ClusterSet referenceClusterSet,
ClusterSet uemClusterSet,
String partialKey,
String method,
ClusterSet clusterSetBase,
ClusterSet clusterSet,
AudioFeatureSet featureSet,
Parameter parameter)
throws Exception {
String oldSpeechDetectorMethod = parameter.getParameterInputFeature().getSpeechMethodAsString();
double oldSpeechDetectorThreshold = parameter.getParameterInputFeature().getSpeechThreshold();
String oldModelKind = parameter.getParameterModel().getModelKindAsString();
int oldNumberOfComponent = parameter.getParameterModel().getNumberOfComponents();
String oldMethod = parameter.getParameterClustering().getMethodAsString();
double oldThreshold = parameter.getParameterClustering().getThreshold();
String oldEMControl = parameter.getParameterEM().getEMControl();
int oldNTop = parameter.getParameterTopGaussian().getScoreNTop();
boolean oldSaveAll = parameter.getParameterDiarization().isSaveAllStep();
DiarizationResultList localResult = new DiarizationResultList(cMin, cMax, mult);
DiarizationError computeError = new DiarizationError(referenceClusterSet, uemClusterSet);
double prevScore = cMin;
// ---- Begin NEW v 1.13 ---
parameter.getParameterInputFeature().setSpeechMethod("E");
parameter.getParameterInputFeature().setSpeechThreshold(0.1);
// ---- End NEW v 1.13 ---
String FeatureFormat = "featureSetTransformation";
String dir = "ester2";
InputStream ubmInputStream = getClass().getResourceAsStream(dir + "/ubm.gmm");
GMMArrayList ubmVect =
MainTools.readGMMContainer(ubmInputStream, parameter.getParameterModel());
GMM ubm = ubmVect.get(0);
/*
* int nbCep = 16; logger.info("---> nbCep:"+nbCep); FeatureSet featureSet2 = loadFeature(parameter, clusterSet, "audio16kHz2sphinx,1:3:2:0:0:0,"+nbCep+",1:1:300:4"); logger.info("---> nbCep:"+nbCep); //logger.fine("*** nbFeaturesNorm:" +
* clusterSetBase.getLength());
*/
// A tester
AudioFeatureSet featureSet2 =
loadFeature(featureSet, parameter, clusterSet, FeatureFormat + ",1:3:2:0:0:0,13,1:1:300:4");
// v5.14
// FeatureSet featureSet2 = loadFeature(featureSet, parameter, clusterSet, FeatureFormat
// + ",1:3:2:0:0:0,13,1:1:0:0");
parameter.getParameterModel().setModelKind("DIAG");
parameter.getParameterModel().setNumberOfComponents(ubm.getNbOfComponents());
parameter.getParameterClustering().setMethod(method);
// ---- Begin NEW v 1.19 ---
// parameter.getParameterEM().setEMControl("1,1,0.01");
// parameter.getParameterClustering().setThreshold(0);
// ---- End NEW v 1.19 ---
parameter.getParameterClustering().setThreshold(cMax);
parameter.getParameterEM().setEMControl("1,5,0.01");
parameter.getParameterTopGaussian().setScoreNTop(5);
parameter.getParameterDiarization().setSaveAllStep(false);
CLRHClustering clustering = new CLRHClustering(clusterSet, featureSet2, parameter, ubm);
// int nbCluster = clusterSet.clusterGetSize();
// logger.info("initialise clustering CLR clusterSet:"+clusterSet);
clustering.initialize();
double score = clustering.getScoreOfCandidatesForMerging();
DiarizationResult error = computeError.scoreOfMatchedSpeakers(clustering.getClusterSet());
double errorRate = error.getErrorRate();
localResult.setResult(prevScore, score, error);
// prevScore = Math.max(score, prevScore);
logger.fine(
"first "
+ parameter.show
+ " key="
+ partialKey
+ " clrScore="
+ score
+ " clrErrorRate="
+ errorRate
+ " clrSize="
+ clustering.getSize()
+ "/"
+ referenceClusterSet.clusterGetSize());
while ((score < cMax) && (clustering.getSize() > 1)) {
localResult.setResult(prevScore, score, error);
prevScore = Math.max(score, prevScore);
clustering.mergeCandidates();
// -- start V5.16 --
// logger.info("--> Decoding");
// ClusterSet decodeClusterSet = MDecode.make(featureSet2, clustering.getClusterSet(),
// clustering.getGmmList(), parameter);
// logger.info("--> Clustering");
// featureSet2 = loadFeature(featureSet, parameter, decodeClusterSet, FeatureFormat
// + ",1:3:2:0:0:0,13,1:1:300:4");
// clustering = new CLRHClustering(decodeClusterSet, featureSet2, parameter, ubm);
// clustering.initialize();
// -- end V5.16 --
score = clustering.getScoreOfCandidatesForMerging();
error = computeError.scoreOfMatchedSpeakers(clustering.getClusterSet());
errorRate = error.getErrorRate();
// localResult.setResult(prevScore, score, error);
// prevScore = Math.max(score, prevScore);
logger.fine(
parameter.show
+ " key="
+ partialKey
+ " clrScore="
+ score
+ " clrErrorRate="
+ errorRate
+ " clrSize="
+ clustering.getSize()
+ "/"
+ referenceClusterSet.clusterGetSize());
}
localResult.setResult(prevScore, score, error);
localResult.setResult(score, cMax, error);
logger.finer(parameter.show + " key=" + partialKey + " resultat du fichier");
localResult.log("partial result: " + parameter.show + " " + partialKey);
clustering.reset();
parameter.getParameterModel().setNumberOfComponents(oldNumberOfComponent);
parameter.getParameterModel().setModelKind(oldModelKind);
parameter.getParameterClustering().setMethod(oldMethod);
parameter.getParameterClustering().setThreshold(oldThreshold);
parameter.getParameterEM().setEMControl(oldEMControl);
parameter.getParameterTopGaussian().setScoreNTop(oldNTop);
parameter.getParameterInputFeature().setSpeechMethod(oldSpeechDetectorMethod);
parameter.getParameterInputFeature().setSpeechThreshold(oldSpeechDetectorThreshold);
parameter.getParameterDiarization().setSaveAllStep(oldSaveAll);
return localResult;
}