Mercurial > hg > multitrack-audio-matcher

changeset 11:9a2b008c4706

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Priors for each channel and scrolling projection for new prior is added
author Andrew N Robertson <andrew.robertson@eecs.qmul.ac.uk>
date Sun, 05 Feb 2012 00:00:58 +0000
parents cbadb9d05d29
children 446121327276
files README.txt bayesianArraySrc/BayesianArrayStructure.cpp bayesianArraySrc/BayesianArrayStructure.h src/AudioEventMatcher.cpp src/AudioEventMatcher.h
diffstat 5 files changed, 88 insertions(+), 73 deletions(-) [+]
line wrap: on
line diff
--- a/README.txt	Sat Feb 04 19:59:27 2012 +0000
+++ b/README.txt	Sun Feb 05 00:00:58 2012 +0000
@@ -33,6 +33,8 @@
 
 OPTIMISE the starting point for the update routine when new`Pitch or onset is found. goes all the way through chromaOnsets
 
+check: Should the offset potion be the alignment position or the actual time in - i guess the former but are we using the latter?
 
 
 
+
--- a/bayesianArraySrc/BayesianArrayStructure.cpp	Sat Feb 04 19:59:27 2012 +0000
+++ b/bayesianArraySrc/BayesianArrayStructure.cpp	Sun Feb 05 00:00:58 2012 +0000
@@ -107,10 +107,19 @@
 	printf("speed adding to index for 1 = %f\n", relativeSpeedPosterior.getRealTermsAsIndex(1));
 	relativeSpeedPosterior.addToIndex(relativeSpeedPosterior.getRealTermsAsIndex(1), 0.1);
 	relativeSpeedPosterior.addGaussianShapeFromRealTime(1, 3, 0.5);
+
+	
+	//tmp debug test
+	//relativeSpeedPosterior.addToIndex(relativeSpeedPosterior.getRealTermsAsIndex(1.8), 0.15);
 	
 	relativeSpeedPosterior.renormalise();
 	relativeSpeedPosterior.getMaximum();
+	
+	
 	relativeSpeedPrior.copyFromDynamicVector(relativeSpeedPosterior);
+	
+	
+	
 	printf("BAYES STRUCTU ' SPEED PRIOR %f . index %i\n", speedPriorValue, index);
 	
 }
@@ -662,8 +671,8 @@
 }
 
 //PROJECT PRIOR CODE
-/*
-void BayesianArrayStructure::projectDistribution(const double& newEventTime, DynamicVector& projectedPrior){
+
+void BayesianArrayStructure::projectDistribution(const double& newEventTime, const double& newAlignmentPosition, DynamicVector& projectedPrior){
 	
 	projectedPrior.copyFromDynamicVector(posterior);
 
@@ -674,89 +683,71 @@
 //	}
 
 	
-	updateBestEstimate(timeDifference);
-	lastBestEstimateUpdateTime = newEventTime;//getTimeNow(timePlayed);
+//	updateBestEstimate(timeDifference);
+//	lastBestEstimateUpdateTime = newEventTime;//getTimeNow(timePlayed);
 	
 	//set TARGETS - commented tenmporarily
-	setNewDistributionOffsets(max(0., bestEstimate - (prior.scalar*prior.arraySize/2)));
-	
-	crossUpdateArrays(posterior, relativeSpeedPosterior, timeDifference);
-	
+
+	//setNewDistributionOffsets(max(0., bestEstimate  - (prior.scalar*prior.arraySize/2)));
+	projectedPrior.offset = posterior.offset;//max(0., newAlignmentPosition - (projectedPrior.scalar*projectedPrior.arraySize/2));
+							
+//	int timeDifference = newEventTime - lastEventTime;
+								
+		double timeDifferenceInPositionVectorUnits = timeDifference / projectedPrior.scalar;
+								
+	//	printf("CROSS UPDATE time diff %f ms is %f units; ", timeDifference, timeDifferenceInPositionVectorUnits);
+		projectedPrior.zero();//kill prior
+								
+								//	calculateNewPriorOffset(timeDifference);//dioesnt do anything
+								
+								//	printf("new prior offset %f and post offset %f\n", prior.offset, posterior.offset);
+								
+				if (timeDifferenceInPositionVectorUnits > crossUpdateTimeThreshold)
+					complexCrossUpdateProjection(projectedPrior, timeDifferenceInPositionVectorUnits);
+				else
+					translatePosteriorByMaximumSpeed(projectedPrior, timeDifferenceInPositionVectorUnits);	
+								
+								
+		//			updateCounter++;
+		projectedPrior.renormalise();//not strictly necessary??
+								
+								
 	//i.e. using the same offset as prior
-	posterior.offset = prior.offset;// 
+//	posterior.offset = prior.offset;// 
 	
 	//	float tmpPrior = max(0., bestEstimate - (prior.scalar*prior.arraySize/2));//	prior.offset = max(0., bestEstimate - (prior.scalar*prior.arraySize/2));
 	//	printf("Using prior offset of %f not %f\n", tmpPrior, prior.offset);
 	
-	lastEventTime = newEventTime;//lastEventTime = ofGetElapsedTimeMillis();
+//	lastEventTime = newEventTime;//lastEventTime = ofGetElapsedTimeMillis();
 	
 }
 
 
 
-
-void BayesianArrayStructure::crossUpdateArrays(DynamicVector& position, DynamicVector& speed, double timeDifference){
-	
-	//set the cutoff for offset of position first! XXX
-	
-	double timeDifferenceInPositionVectorUnits = timeDifference / prior.scalar;
-	
-	printf("CROSS UPDATE time diff %f ms is %f units; ", timeDifference, timeDifferenceInPositionVectorUnits);
-	prior.zero();//kill prior
-	
-	//	calculateNewPriorOffset(timeDifference);//dioesnt do anything
-	
-	//	printf("new prior offset %f and post offset %f\n", prior.offset, posterior.offset);
-	
-	if (timeDifferenceInPositionVectorUnits > crossUpdateTimeThreshold)
-		complexCrossUpdate(timeDifferenceInPositionVectorUnits);
-	else
-		translateByMaximumSpeed(timeDifferenceInPositionVectorUnits);	
-	
-	
-	updateCounter++;
-	prior.renormalise();//not strictly necessary??
-	
-}
-
-void BayesianArrayStructure::complexCrossUpdate(const double& timeDifferenceInPositionVectorUnits){
-	
-	
-	printf("before cross c : posterior map is %i = %f ms time diff pos vec %f\n", posterior.getMAPestimate(), posterior.getIndexInRealTerms(prior.getMAPestimate()), timeDifferenceInPositionVectorUnits);
+void BayesianArrayStructure::complexCrossUpdateProjection(DynamicVector& projectedPrior, const double& timeDifferenceInPositionVectorUnits){
 	
 	int distanceMoved, newPriorIndex;
-	
+
 	double speedValue = relativeSpeedPosterior.offset;
 	
 	for (int i = 0;i < relativeSpeedPosterior.arraySize;i++){
 		
-		//	double speedValue = relativeSpeedPosterior.getIndexInRealTerms(i);//so for scalar 0.01, 50 -> speed value of 0.5
-		
-		//so we have moved 
 		distanceMoved = round(timeDifferenceInPositionVectorUnits * speedValue);//round the value
-		//	printf("Speed value %f time %f gives distance %i\n", speedValue, timeDifferenceInPositionVectorUnits, distanceMoved);
 		
 		if (relativeSpeedPosterior.array[i] != 0){
 			
 			double speedContribution = relativeSpeedPosterior.array[i];
 			
-			//	printf("speed [%i](val[%f]) gives %f moved %i in %f units \n", i, relativeSpeedPosterior.array[i], speedValue, distanceMoved, timeDifferenceInPositionVectorUnits);
-			
-			//1/2/12 deleted line
-			newPriorIndex = posterior.offset - prior.offset + distanceMoved;//i.e. where post[0] goes to in terms of prior at this speed
+			newPriorIndex = posterior.offset - projectedPrior.offset + distanceMoved;//i.e. where post[0] goes to in terms of prior at this speed
 			int postIndex = 0;//index of posterior that will contribute
 			
-			while (postIndex < posterior.arraySize && newPriorIndex < prior.arraySize){
-				
-				//did use a for loop
-				//	for (postIndex = 0;postIndex < posterior.arraySize;postIndex++){
-				//old posterior contributing to new prior
-				
+			while (postIndex < posterior.arraySize && newPriorIndex < projectedPrior.arraySize){
+
 				//would use this method
 				//newPriorIndex = postIndex + posterior.offset - prior.offset + distanceMoved;
 				
 				if (newPriorIndex >= 0){
-					prior.addToIndex(newPriorIndex, posterior.array[postIndex]*speedContribution);
+					projectedPrior.addToIndex(newPriorIndex, posterior.array[postIndex]*speedContribution);
 					//		printf("speed index %i new prior index %i post val %f speed contrib %f dist %i\n", i, newPriorIndex, posterior.array[postIndex], speedContribution, distanceMoved);
 				}
 				//but we actually do this for simplicity
@@ -772,14 +763,12 @@
 		//	double speedValue = relativeSpeedPosterior.getIndexInRealTerms(i);//so for scalar 0.01, 50 -> speed value of 0.5
 		
 	}//end speed
-	
-	
-	printf("after cross c : prior map is %i = %f ms\n", prior.getMAPestimate(), prior.getIndexInRealTerms(prior.getMAPestimate()));
+
 }
 
 
 
-void BayesianArrayStructure::translateByMaximumSpeed(const double& timeDifferenceInPositionVectorUnits){
+void BayesianArrayStructure::translatePosteriorByMaximumSpeed(DynamicVector& translatedPosterior, const double& timeDifferenceInPositionVectorUnits){
 	
 	
 	int distanceMoved, newPriorIndex;
@@ -799,17 +788,18 @@
 		//old posterior contributing to new prior
 		newPriorIndex = postIndex + posterior.offset - prior.offset + distanceMoved;
 		if (newPriorIndex >= 0 && newPriorIndex < prior.arraySize){
-			prior.addToIndex(newPriorIndex, posterior.array[postIndex]*speedContribution);
+			translatedPosterior.addToIndex(newPriorIndex, posterior.array[postIndex]*speedContribution);
 		}
 		
 	}
 	
 }
- */
+
 
 //END PROJECT PRIOR CODE
 
 
+
 /*
  
  void BayesianArrayStructure::updateTempoDistribution(const double& speedRatio, const double& matchFactor){
--- a/bayesianArraySrc/BayesianArrayStructure.h	Sat Feb 04 19:59:27 2012 +0000
+++ b/bayesianArraySrc/BayesianArrayStructure.h	Sun Feb 05 00:00:58 2012 +0000
@@ -98,6 +98,10 @@
 	bool usingIntegratedTempoEstimate;
 	double relativeSpeedLikelihoodStdDev;
 	
+	void projectDistribution(const double& newEventTime, const double& newAlignmentPosition, DynamicVector& projectedPrior);
+	void complexCrossUpdateProjection(DynamicVector& projectedPrior, const double& timeDifferenceInPositionVectorUnits);
+	void translatePosteriorByMaximumSpeed(DynamicVector& translatedPosterior, const double& timeDifferenceInPositionVectorUnits);
+	
 	void printPostOffset();
 	bool updatingSpeedDistribution;//false for testing
 
--- a/src/AudioEventMatcher.cpp	Sat Feb 04 19:59:27 2012 +0000
+++ b/src/AudioEventMatcher.cpp	Sun Feb 05 00:00:58 2012 +0000
@@ -51,6 +51,8 @@
 	bayesianStruct.setStartPlaying();
 	currentAlignmentPosition = 0;
 	startTime = ofGetElapsedTimeMillis();
+	
+	projectedPrior = bayesianStruct.prior;
 	//bayesianStruct.posterior.printArray();
 }
 
@@ -72,10 +74,12 @@
 //	double tmp = bayesianStruct.posterior.getIndexInRealTerms(bayesianStruct.posterior.MAPestimate);;
 //	double timetmp = (newTime - lastAlignmentTime);
 //	double speedtmp = bayesianStruct.relativeSpeedPosterior.getIndexInRealTerms(bayesianStruct.relativeSpeedPosterior.MAPestimate);
-	
+//	currentAlignmentTime = newTime;
 	currentAlignmentPosition = bayesianStruct.posterior.getIndexInRealTerms(bayesianStruct.posterior.MAPestimate);
 	currentAlignmentPosition += (newTime - lastAlignmentTime) * bayesianStruct.relativeSpeedPosterior.getIndexInRealTerms(bayesianStruct.relativeSpeedPosterior.MAPestimate);
 
+	bayesianStruct.projectDistribution(newTime, currentAlignmentPosition, projectedPrior);//prior gets updated to where we are now
+	
 //	printf("ALIGN pos %f time diff %f (now %f , last %f)speed %f :: ALIGN BEST %f\n", tmp, timetmp, (double)ofGetElapsedTimeMillis(), lastAlignmentTime, speedtmp, currentAlignmentPosition);
 }
 
@@ -88,19 +92,15 @@
 	//draw the scrolling audio tracks
 	recordedTracks.drawTracks();
 
-
-	
 	ofSetColor(255);
 //	bayesianStruct.relativeSpeedPrior.drawVector(0, 200, bayesTempoWindow);
 
 	setScreenDisplayTimes();
 	drawBayesianDistributions();
 	
-	//	bayesianStruct.posterior.drawVector(0, bayesianStruct.posterior.getRealTermsAsIndex(screenWidthMillis), bayesPositionWindow);
-	
+	//bayesianStruct.posterior.drawVector(0, bayesianStruct.posterior.getRealTermsAsIndex(screenWidthMillis), bayesPositionWindow);
 	//bayesianStruct.posterior.drawVector(bayesianStruct.posterior.getRealTermsAsIndex(0), bayesianStruct.posterior.getRealTermsAsIndex(screenWidthMillis), bayesPositionWindow);
-	
-	//	bayesianStruct.relativeSpeedPosterior.drawVector(0, bayesianStruct.relativeSpeedPosterior.getRealTermsAsIndex(2), bayesTempoWindow);
+	//bayesianStruct.relativeSpeedPosterior.drawVector(0, bayesianStruct.relativeSpeedPosterior.getRealTermsAsIndex(2), bayesTempoWindow);
 	
 	ofDrawBitmapString("pitch "+ofToString(recentPitch, 2)+", Time "+ofToString(recentTime, 0), 20, 20);
 	
@@ -157,6 +157,11 @@
 	for (int i = 0; i <recordedTracks.numberOfAudioTracks;i++){
 		ofSetColor(200,255,50);
 		likelihoodVisualisation[i].drawConstrainedVector(likelihoodVisualisation[i].getRealTermsAsIndex(screenStartTimeMillis), likelihoodVisualisation[i].getRealTermsAsIndex(screenEndTimeMillis), 0, ofGetWidth(), recordedTracks.loadedAudioFiles[i].fileLoader.onsetDetect.window);		
+		
+		ofSetColor(0,255,150);
+		recentPriors[i].drawConstrainedVector(recentPriors[i].getRealTermsAsIndex(screenStartTimeMillis), recentPriors[i].getRealTermsAsIndex(screenEndTimeMillis), 0, ofGetWidth(), recordedTracks.loadedAudioFiles[i].fileLoader.onsetDetect.window);		
+		
+		
 		ofSetColor(255);
 		ofDrawBitmapString("recent event "+ofToString(recentEventTime[i]), recordedTracks.loadedAudioFiles[i].fileLoader.onsetDetect.window.x + 20, recordedTracks.loadedAudioFiles[i].fileLoader.onsetDetect.window.y + recordedTracks.loadedAudioFiles[i].fileLoader.onsetDetect.window.height - 10);
 	}
@@ -167,7 +172,10 @@
 	recentPrior.drawConstrainedVector(priorStartIndex, priorEndIndex, 0, ofGetWidth(), bayesPositionWindow);
 
 	ofSetColor(255,0,100);//purple prior
-	bayesianStruct.prior.drawConstrainedVector(bayesianStruct.prior.getRealTermsAsIndex(screenStartTimeMillis), bayesianStruct.prior.getRealTermsAsIndex(screenEndTimeMillis), 0, ofGetWidth(), bayesLikelihoodWindow);
+	bayesianStruct.prior.drawConstrainedVector(bayesianStruct.prior.getRealTermsAsIndex(screenStartTimeMillis), bayesianStruct.prior.getRealTermsAsIndex(screenEndTimeMillis), 0, ofGetWidth(), bayesPositionWindow);
+	
+	ofSetColor(255,0,0);
+	projectedPrior.drawConstrainedVector(bayesianStruct.prior.getRealTermsAsIndex(screenStartTimeMillis), bayesianStruct.prior.getRealTermsAsIndex(screenEndTimeMillis), 0, ofGetWidth(), bayesLikelihoodWindow);
 	
 }
 
@@ -176,7 +184,7 @@
 	liveInput.addPitchEvent(pitchIn, timeIn);
 	
 	//printPosteriorMAPinfo();	
-		
+
 	matchNewPitchEvent(channel, pitchIn, timeIn);//main pitch matching fn
 
 	likelihoodVisualisation[1] = bayesianStruct.likelihood;
@@ -243,14 +251,20 @@
 		}
 	}
 	
+	if (numberOfMatchesFound > 0){
 //	bayesianStruct.likelihood.addConstant((1-likelihoodToNoiseRatio)/bayesianStruct.likelihood.length);
 	bayesianStruct.likelihood.addConstant(numberOfMatchesFound*(1-onsetLikelihoodToNoise)/(onsetLikelihoodToNoise*bayesianStruct.likelihood.length));
 	bayesianStruct.likelihood.renormalise();
 	
 	bayesianStruct.calculatePosterior();
-	
 	lastAlignmentTime = timeIn;//use TIMESTAMP
 	recentEventTime[channel] = timeIn;//ofGetElapsedTimeMillis() - startTime;
+
+	recentPriors[channel] = bayesianStruct.prior;
+	
+	}
+	
+
 	
 }
 
@@ -300,10 +314,13 @@
 	//bayesianStruct.likelihood.addConstant(1);
 		
 		bayesianStruct.calculatePosterior();
+		lastAlignmentTime = timeIn;//has to use the STAMPED time
+		recentEventTime[channel] = timeIn;
+		
+		recentPriors[channel] = bayesianStruct.prior;
 	}
 	
-	lastAlignmentTime = timeIn;//has to use the STAMPED time
-	recentEventTime[channel] = timeIn;
+
 }
 
 double AudioEventMatcher::getPitchDistance(const double& pitchOne, const double& pitchTwo, const double& scale){
--- a/src/AudioEventMatcher.h	Sat Feb 04 19:59:27 2012 +0000
+++ b/src/AudioEventMatcher.h	Sun Feb 05 00:00:58 2012 +0000
@@ -64,6 +64,7 @@
 	double recentPitch, recentTime;
 	
 	DynamicVector likelihoodVisualisation[3];
+	DynamicVector recentPriors[3];
 	DynamicVector recentPrior;
 	DynamicVector projectedPrior;
 	
@@ -72,6 +73,7 @@
 	
 	double recentEventTime[3];
 	int startTime;
+	int currentAlignmentTime;
 	
 	double screenStartTimeMillis, screenEndTimeMillis, screenWidthMillis;
 	bool followingLiveInput;