Mercurial > hg > multitrack-audio-matcher

changeset 37:9806a4f22fd0

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Fixed bugs in the likelohoods that caused some to zero when no events found.
author Andrew N Robertson <andrew.robertson@eecs.qmul.ac.uk>
date Tue, 24 Apr 2012 14:16:01 +0100
parents eb43b2a007ea
children 9c803369b229
files bayesianArraySrc/BayesianArrayStructure.cpp src/AudioEventMatcher.cpp src/AudioEventMatcher.h src/LiveAudioInput.cpp src/TempoFollower.cpp src/testApp.cpp
diffstat 6 files changed, 79 insertions(+), 38 deletions(-) [+]
line wrap: on
line diff
--- a/bayesianArraySrc/BayesianArrayStructure.cpp	Tue Apr 24 01:19:24 2012 +0100
+++ b/bayesianArraySrc/BayesianArrayStructure.cpp	Tue Apr 24 14:16:01 2012 +0100
@@ -31,7 +31,7 @@
 
 	tmpBestEstimate = 0;
 	crossUpdateTimeThreshold = 60;
-	priorWidth = 12;
+	priorWidth = 20;
 	
 
 	
@@ -106,7 +106,9 @@
 	speedPriorValue = f;
 	int index = relativeSpeedPosterior.getRealTermsAsIndex(speedPriorValue);
 	relativeSpeedPosterior.zero();
-	relativeSpeedPosterior.addGaussianShape(index, priorWidth, 0.9);
+	
+	//changed these a bit - still need to figure tempo process out properly
+	relativeSpeedPosterior.addGaussianShape(index, priorWidth, 1.9);
 	relativeSpeedPosterior.addGaussianShape(relativeSpeedPosterior.getRealTermsAsIndex(1), 4, 1);
 	
 	printf("speed adding to index for 1 = %f\n", relativeSpeedPosterior.getRealTermsAsIndex(1));
--- a/src/AudioEventMatcher.cpp	Tue Apr 24 01:19:24 2012 +0100
+++ b/src/AudioEventMatcher.cpp	Tue Apr 24 14:16:01 2012 +0100
@@ -17,6 +17,8 @@
 
 	useChromaDotProduct = false;
 	
+	printingData = false;
+	
 	pitchLikelihoodToNoise = 0.6;//more noise
 	chromaLikelihoodToNoise = 0.5;//lower => more noise, higher more weight for events
 	chromaLikelihoodWidth = 50;//ms round onset event
@@ -146,6 +148,9 @@
 	//SET TEMPO PRIOR for Speed Ratio
 	//the update this
 	setSpeedRatioDistribution(currentSpeedRatio);
+	
+	euclideanMaximumDistance = 0;
+	
 	//bayesianStruct.posterior.printArray();
 }
 
@@ -158,7 +163,7 @@
 
 void AudioEventMatcher::stopPlaying(){
 	startedPlaying = false;
-	temporal.printEventTimes();
+	//temporal.printEventTimes();
 }
 
 void AudioEventMatcher::rescue(){
@@ -186,6 +191,7 @@
 
 void AudioEventMatcher::updateRecordedTempo(){
 	//tempo of equivalent recorded position is updated 
+	if (recordedTempoIndex < recordedTempoData.globalTempoTimes.size()){//if for debug
 	while(currentAlignmentPosition > recordedTempoData.globalTempoTimes[recordedTempoIndex]){
 		recordedTempoIndex++;
 	}
@@ -194,6 +200,8 @@
 	currentSpeedRatio = temporal.playingTempo / recordedTempo;
 	if (currentSpeedRatio != tmpRatio)
 		setSpeedRatioDistribution(currentSpeedRatio);
+	
+	}//end if to prevent debug crash
 }
 
 void AudioEventMatcher::updateBestAlignmentPosition(){
@@ -343,10 +351,18 @@
 	ofSetColor(255,0,0);//projected prior in red
 	projectedPrior.drawConstrainedVector(bayesianStruct.prior.getRealTermsAsIndex(screenStartTimeMillis), bayesianStruct.prior.getRealTermsAsIndex(screenEndTimeMillis), 0, ofGetWidth(), bayesPositionWindow);
 	
+	//draw pitch
+	ofSetColor(0,100,255);
+	int index = getScreenWidthIndexOfEventTime(recentPitchEventTime);
+	//this window would be used (recordedTracks.loadedAudioFiles[1].fileLoader.onsetDetect.window);
 	
 	
 }
 
+int AudioEventMatcher::getScreenWidthIndexOfEventTime(const double& time){
+	return (time - screenStartTimeMillis)*ofGetWidth()/screenWidthMillis;
+}
+
 void AudioEventMatcher::drawTrackLikelihoods(){
 	//draw track by track likelihoods
 	for (int i = 0; i <recordedTracks.numberOfAudioTracks;i++){
@@ -376,7 +392,7 @@
 	tmpStr = "pitch "+ofToString(recentPitch, 2);
 	tmpStr += " Nearest "+ofToString(pitchOfNearestMatch,2);
 	tmpStr += " dist "+ofToString(distanceOfNearestMatch, 2);
-	tmpStr +=  ", Time "+ofToString(recentTime, 0);
+	tmpStr +=  ", Time "+ofToString(recentPitchEventTime, 0);
 	ofDrawBitmapString(tmpStr, 20, 20);
 	
 	string alignString = " align "+ofToString(currentAlignmentPosition, 2);
@@ -400,7 +416,7 @@
 	likelihoodVisualisation[1] = bayesianStruct.likelihood;
 		
 	recentPitch = pitchIn;//for drawing
-	recentTime = timeIn;
+	recentPitchEventTime = timeIn;
 	}
 }
 
@@ -408,9 +424,11 @@
 void AudioEventMatcher::newChromaEvent(const int& channel, float* chromaIn, const double& timeIn){
 	
 	//	could add event to the liveInput list? as in pitch event
-	printf("match chroma channel %i\n", channel);	
-	for (int i = 0;i < 12;i++){
+	if (printingData){
+		printf("match chroma channel %i\n", channel);	
+		for (int i = 0;i < 12;i++){
 		printf("chroma in[%i] = %f\n", i, chromaIn[i]);
+		}
 	}
 	
 	matchNewChromaEvent(channel, chromaIn, timeIn);//main pitch matching fn
@@ -544,9 +562,11 @@
 
 	
 	
-	if (numberOfMatches > 0){//no point updating unless there is a match
+	if (numberOfMatches > 0 && totalLikelihoodAdded > 0){//no point updating unless there is a match
 	//replacing numberOfMatches with totalLike below...
-		bayesianStruct.likelihood.addConstant(totalLikelihoodAdded*(1-pitchLikelihoodToNoise)/(pitchLikelihoodToNoise*bayesianStruct.likelihood.length));
+		//bug here was that if totaladded = 0, we add then zero likelihood
+		bayesianStruct.likelihood.addConstant(totalLikelihoodAdded*(1-pitchLikelihoodToNoise)/(bayesianStruct.likelihood.length));
+	//	bayesianStruct.likelihood.addConstant(totalLikelihoodAdded*(1-pitchLikelihoodToNoise)/(pitchLikelihoodToNoise*bayesianStruct.likelihood.length));
 
 	//tmp set likelihood constant and calculate using that
 	//bayesianStruct.likelihood.zero();
@@ -622,7 +642,8 @@
 	double millisTime = -1*INFINITY;//or 0 is fine
 	
 	int checkIndex = 0;
-	if (channel <= recordedTracks.numberOfAudioTracks){
+	if (channel <= recordedTracks.numberOfAudioTracks && checkIndex < recordedTracks.loadedAudioFiles[channel].fileLoader.onsetDetect.chromaOnsets.size()){
+
 		while (millisTime < startMatchingTime) {
 			millisTime = recordedTracks.loadedAudioFiles[channel].fileLoader.onsetDetect.chromaOnsets[checkIndex].millisTime;
 			checkIndex++;
@@ -644,19 +665,22 @@
 			//	bayesianStruct.likelihood.addGaussianShapeFromRealTime(millisTime, onsetLikelihoodWidth,  quantity);
 				numberOfMatches++;
 				totalLikelihoodAdded += quantity;
-				printf("Adding CHROMA Gaussian for onset at time %.1f dist %.3f\n", millisTime, quantity);
+				
+				//printf("Adding CHROMA Gaussian for onset at time %.1f dist %.3f\n", millisTime, quantity);
 				
 			}//end if within limits (changed so it now is 4 sure)
 		}
 	}
 	
 	
-	if (numberOfMatches > 0){//no point updating unless there is a match
+	if (numberOfMatches > 0 && totalLikelihoodAdded > 0){//no point updating unless there is a match
 		//replacing numberOfMatches with totalLike below...
 		
 		printf("CHROMA HAS %i MATCHES\n", numberOfMatches);
 		
-		bayesianStruct.likelihood.addConstant(totalLikelihoodAdded*(1-chromaLikelihoodToNoise)/(chromaLikelihoodToNoise*bayesianStruct.likelihood.length));
+		bayesianStruct.likelihood.addConstant(totalLikelihoodAdded*(1-chromaLikelihoodToNoise)/(bayesianStruct.likelihood.length));
+	//previous way	
+	//	bayesianStruct.likelihood.addConstant(totalLikelihoodAdded*(1-chromaLikelihoodToNoise)/(chromaLikelihoodToNoise*bayesianStruct.likelihood.length));
 		
 		bayesianStruct.calculatePosterior();
 		lastAlignmentTime = timeIn;//has to use the STAMPED time
@@ -688,6 +712,7 @@
 double AudioEventMatcher::getChromaEuclideanDistance(float* chromaOne, float* chromaTwo){
 	double distance = 0;
 	double total = 0;
+	
 //	printf("\n");
 	for (int i = 0;i < 12;i++){
 		total += (chromaOne[i] - chromaTwo[i])*(chromaOne[i] - chromaTwo[i]);
@@ -695,8 +720,15 @@
 	//	total += chromaOne[i]*chromaOne[i] + (chromaTwo[i]*chromaTwo[i]);
 	}
 	
-	if (total > 0)
-		distance = 1.0/sqrt(total);
+	if (total > euclideanMaximumDistance)
+		euclideanMaximumDistance = total;
+	
+	distance = ((euclideanMaximumDistance - total)/ euclideanMaximumDistance);//i.e. 1 is 
+	
+//	if (total > 0)
+		
+		
+//		distance = 1.0/sqrt(total);
 //	printf("DISTANCE : %.3f\n", distance);
 	return distance;
 }
--- a/src/AudioEventMatcher.h	Tue Apr 24 01:19:24 2012 +0100
+++ b/src/AudioEventMatcher.h	Tue Apr 24 14:16:01 2012 +0100
@@ -43,6 +43,7 @@
 	void drawInfo();
 	
 	void setWindowDimensions();
+	int getScreenWidthIndexOfEventTime(const double& time);
 	
 	void newPitchEvent(const int& channel, const double& pitchIn, const double& timeIn);
 	void newKickEvent(const double& timeIn);
@@ -82,7 +83,7 @@
 	bool startedPlaying;
 	
 	bool usingRealTime;
-	double recentPitch, recentTime;
+	double recentPitch, recentPitchEventTime;
 	
 	DynamicVector likelihoodVisualisation[numberOfChannels];
 	DynamicVector recentPriors[numberOfChannels];
@@ -126,5 +127,8 @@
 	double currentSpeedRatio;
 	void drawPlayingTempo();
 	void setSpeedRatioDistribution(const double& speedRatio);
+	
+	double euclideanMaximumDistance;
+	bool printingData;
 };
 #endif
--- a/src/LiveAudioInput.cpp	Tue Apr 24 01:19:24 2012 +0100
+++ b/src/LiveAudioInput.cpp	Tue Apr 24 14:16:01 2012 +0100
@@ -9,6 +9,10 @@
 
 #include "LiveAudioInput.h"
 
+//Keeping track of incoming events
+//more useful as data than anything else. But in fact, not particularly useful it seems at present
+
+
 const double hopsize = 512;//check to see it agrees with ofxAubioOnsetDetector
 
 LiveAudioInput::LiveAudioInput(){
@@ -16,9 +20,6 @@
 	
 }
 
-
-
-
 void LiveAudioInput::addPitchEvent(const double& pitch, const double& time){
 
 	AudioEvent e;
@@ -32,22 +33,6 @@
 }
 
 
-
-/*
- //any point in this??
-void LiveAudioInput::addChromaEvent(const double& time){
-	
-	AudioEvent e;
-	e.millisTime = time;
-	e.frameTime = millisToFrames(time);
-	
-	liveEvents.push_back(e);
-	//printf("live input pitch %f time %f ms == %f frames\n", pitch, time, e.frameTime);
-	numberOfEvents++;
-	
-}
-*/
-
 double LiveAudioInput::framesToMillis(const double& frameCount){
 	return ((frameCount*hopsize*1000.0)/44100.0);
 }
@@ -55,4 +40,21 @@
 
 double LiveAudioInput::millisToFrames(const double& millis){
 	return ((millis*44100.0)/(hopsize*1000.0));
-}
\ No newline at end of file
+}
+
+
+
+/*
+ //any point in this??
+ void LiveAudioInput::addChromaEvent(const double& time){
+ 
+ AudioEvent e;
+ e.millisTime = time;
+ e.frameTime = millisToFrames(time);
+ 
+ liveEvents.push_back(e);
+ //printf("live input pitch %f time %f ms == %f frames\n", pitch, time, e.frameTime);
+ numberOfEvents++;
+ 
+ }
+ */
\ No newline at end of file
--- a/src/TempoFollower.cpp	Tue Apr 24 01:19:24 2012 +0100
+++ b/src/TempoFollower.cpp	Tue Apr 24 14:16:01 2012 +0100
@@ -117,7 +117,7 @@
 			double divisionInEighthNotes = intervalsToTest[k];
 			double testInterval = interval / divisionInEighthNotes;
 			if (testTempoInterval(channel, testInterval, d)){
-				printf("channel %i interval %i at division %.0f == tempo update %f\n", channel, interval, divisionInEighthNotes, testInterval);
+				//printf("channel %i interval %i at division %.0f == tempo update %f\n", channel, interval, divisionInEighthNotes, testInterval);
 				div.push_back((int)divisionInEighthNotes);
 			}
 		}
--- a/src/testApp.cpp	Tue Apr 24 01:19:24 2012 +0100
+++ b/src/testApp.cpp	Tue Apr 24 14:16:01 2012 +0100
@@ -83,10 +83,11 @@
 			int testChannel = m.getArgAsInt32(0);
 			double timeIn = m.getArgAsFloat(1);
 			float chromaIn[12];
+			
 			printf("CHROMA received at time %f\n", timeIn);
 			for (int i = 0;i < 12;i++){
 				chromaIn[i] = m.getArgAsFloat(i+2);
-				printf("chroma[%i]: %f\n", i, chromaIn[i]);
+			//	printf("chroma[%i]: %f\n", i, chromaIn[i]);
 			}
 			eventMatcher.newChromaEvent(testChannel, chromaIn, timeIn);
 		}