KeyFinderResultWrapper keyDetectionProcess(const AsyncFileObject& object){
KeyFinderResultWrapper result;
result.batchRow = object.batchRow;
// initialise stream and decode file into it.
KeyFinder::AudioData* audio = NULL;
AudioFileDecoder* decoder = NULL;
try{
decoder = AudioFileDecoder::getDecoder();
}catch(KeyFinder::Exception& e){
delete decoder;
result.errorMessage = QString(e.what().c_str());
return result;
}
try{
audio = decoder->decodeFile(object.filePath);
delete decoder;
}catch(KeyFinder::Exception& e){
delete audio;
delete decoder;
result.errorMessage = QString(e.what().c_str());
return result;
}
// make audio stream monaural ahead of downsample to reduce load
audio->reduceToMono();
// downsample if necessary
if(object.prefs.getDFactor() > 1){
Downsampler* ds = Downsampler::getDownsampler(object.prefs.getDFactor(),audio->getFrameRate(),object.prefs.getLastFreq());
try{
audio = ds->downsample(audio,object.prefs.getDFactor());
}catch(KeyFinder::Exception& e){
delete audio;
delete ds;
result.errorMessage = QString(e.what().c_str());
return result;
}
delete ds;
}
KeyFinder::KeyFinder* kf = LibKeyFinderSingleton::getInstance()->getKeyFinder();
result.core = kf->findKey(*audio, object.prefs.core);
delete audio;
return result;
}
int main(int argc, char* argv[]) {
FILE* fin;
FILE* fout;
SDownsamplerCtx sDownsamplerCtx;
SDownsamplerCtx* pDownsamplerCtx = &sDownsamplerCtx;
Downsampler* pDownsampler = NULL;
int iMaxFrameNum = 0;
int iFileSize = 0;
int iFrameSize = 0;
memset(&sDownsamplerCtx, 0, sizeof(sDownsamplerCtx));
parseCmdLine(argc, argv, pDownsamplerCtx);
ChooseDownsampler(pDownsamplerCtx, &pDownsampler);
fin = fopen((const char*)pDownsamplerCtx->pInfileName, "rb");
if(!fin) {
cout << "Error: can not open input file!" << endl;
exit(1);
}
fseek(fin, 0L, SEEK_END);
iFileSize = (int)ftell(fin);
fseek(fin, 0L, SEEK_SET);
fout = fopen((const char*)pDownsamplerCtx->pOutfileName, "wb");
if(!fout) {
cout << "Error: can not open output file!" << endl;
exit(1);
}
//caculate max frame number
iFrameSize = pDownsamplerCtx->iSrcWidth * pDownsamplerCtx->iSrcHeight * 1.5;
iMaxFrameNum = iFileSize / iFrameSize;
pDownsampler->InitDownsampler(pDownsamplerCtx);
for(int i = 0; i < iMaxFrameNum; i++) {
pDownsampler->ReadFrameFromFile(fin);
pDownsampler->Processing();
pDownsampler->WriteFrameToFile(fout);
}
//downsample end
delete pDownsampler;
fclose(fin);
fclose(fout);
return 0;
}
KeyDetectionResult KeyFinder::findKey(const AudioData& originalAudio, const Parameters& params){
KeyDetectionResult result;
AudioData* workingAudio = new AudioData(originalAudio);
workingAudio->reduceToMono();
Downsampler ds;
ds.downsample(workingAudio, params.getLastFreq(), &lpfFactory);
SpectrumAnalyser* sa = new SpectrumAnalyser(workingAudio->getFrameRate(), params, &ctFactory);
// run spectral analysis
Chromagram* ch = sa->chromagram(workingAudio);
delete workingAudio;
delete sa;
// reduce chromagram
ch->reduceTuningBins(params);
result.fullChromagram = Chromagram(*ch);
ch->reduceToOneOctave(params);
result.oneOctaveChromagram = Chromagram(*ch);
// get harmonic change signal
Segmentation* segmenter = Segmentation::getSegmentation(params);
result.harmonicChangeSignal = segmenter->getRateOfChange(*ch, params);
// get track segmentation
std::vector<unsigned int> segmentBoundaries = segmenter->getSegments(result.harmonicChangeSignal, params);
segmentBoundaries.push_back(ch->getHops()); // sentinel
delete segmenter;
// get key estimates for each segment
KeyClassifier hc(params);
std::vector<float> keyWeights(24); // TODO: not ideal using int cast of key_t enum. Hash?
for (int s = 0; s < (signed)segmentBoundaries.size()-1; s++){
KeyDetectionSegment segment;
segment.firstWindow = segmentBoundaries[s];
segment.lastWindow = segmentBoundaries[s+1] - 1;
// collapse segment's time dimension, for a single chroma vector and a single energy value
std::vector<float> segmentChroma(ch->getBins());
// for each relevant hop of the chromagram
for (unsigned int hop = segment.firstWindow; hop <= segment.lastWindow; hop++) {
// for each bin
for (unsigned int bin = 0; bin < ch->getBins(); bin++) {
float value = ch->getMagnitude(hop, bin);
segmentChroma[bin] += value;
segment.energy += value;
}
}
segment.key = hc.classify(segmentChroma);
if(segment.key != SILENCE){
keyWeights[segment.key] += segment.energy;
}
result.segments.push_back(segment);
}
delete ch;
// get global key
result.globalKeyEstimate = SILENCE;
float mostCommonKeyWeight = 0.0;
for (int k = 0; k < (signed)keyWeights.size(); k++){
if(keyWeights[k] > mostCommonKeyWeight){
mostCommonKeyWeight = keyWeights[k];
result.globalKeyEstimate = (key_t)k;
}
}
return result;
}
void KeyFinderWorkerThread::run(){
if(!haveParams){
emit failed("No parameters.");
return;
}
// initialise stream and decode file into it
AudioStream* astrm = NULL;
AudioFileDecoder* dec = AudioFileDecoder::getDecoder(filePath.toUtf8().data());
try{
astrm = dec->decodeFile(filePath.toUtf8().data());
}catch(Exception){
delete astrm;
delete dec;
emit failed("Could not decode file.");
return;
}
delete dec;
emit decoded();
// make audio stream monaural
astrm->reduceToMono();
emit madeMono();
// downsample if necessary
if(prefs.getDFactor() > 1){
Downsampler* ds = Downsampler::getDownsampler(prefs.getDFactor(),astrm->getFrameRate(),prefs.getLastFreq());
try{
astrm = ds->downsample(astrm,prefs.getDFactor());
}catch(Exception){
delete astrm;
delete ds;
emit failed("Downsampler failed.");
return;
}
delete ds;
emit downsampled();
}
// start spectrum analysis
SpectrumAnalyser* sa = NULL;
Chromagram* ch = NULL;
sa = SpectrumAnalyserFactory::getInstance()->getSpectrumAnalyser(astrm->getFrameRate(),prefs);
ch = sa->chromagram(astrm);
delete astrm; // note we don't delete the spectrum analyser; it stays in the centralised factory for reuse.
ch->reduceTuningBins(prefs);
emit producedFullChromagram(*ch);
// reduce chromagram
ch->reduceToOneOctave(prefs);
emit producedOneOctaveChromagram(*ch);
// get energy level across track to weight segments
std::vector<float> loudness(ch->getHops());
for(int h=0; h<ch->getHops(); h++)
for(int b=0; b<ch->getBins(); b++)
loudness[h] += ch->getMagnitude(h,b);
// get harmonic change signal
Segmentation* hcdf = Segmentation::getSegmentation(prefs);
std::vector<double> harmonicChangeSignal = hcdf->getRateOfChange(ch,prefs);
emit producedHarmonicChangeSignal(harmonicChangeSignal);
// get track segmentation
std::vector<int> changes = hcdf->getSegments(harmonicChangeSignal,prefs);
changes.push_back(ch->getHops()); // It used to be getHops()-1. But this doesn't crash. So we like it.
// batch output of keychange locations for Beatles experiment
//for(int i=1; i<changes.size(); i++) // don't want the leading zero
// std::cout << filePath.substr(53) << "\t" << std::fixed << std::setprecision(2) << changes[i]*(prefs.getHopSize()/(44100.0/prefs.getDFactor())) << std::endl;
// end experiment output
// get key estimates for segments
KeyClassifier hc(prefs);
std::vector<int> keys(0);
std::vector<float> keyWeights(24);
for(int i=0; i<(signed)changes.size()-1; i++){
std::vector<double> chroma(ch->getBins());
for(int j=changes[i]; j<changes[i+1]; j++)
for(int k=0; k<ch->getBins(); k++)
chroma[k] += ch->getMagnitude(j,k);
int key = hc.classify(chroma);
for(int j=changes[i]; j<changes[i+1]; j++){
keys.push_back(key);
if(key < 24) // ignore parts that were classified as silent
keyWeights[key] += loudness[j];
}
}
keys.push_back(keys[keys.size()-1]); // put last key on again to match length of track
delete ch;
emit producedKeyEstimates(keys);
// get global key
int mostCommonKey = 24;
float mostCommonKeyWeight = 0.0;
for(int i=0; i<(signed)keyWeights.size(); i++){
if(keyWeights[i] > mostCommonKeyWeight){
mostCommonKeyWeight = keyWeights[i];
mostCommonKey = i;
}
}
emit producedGlobalKeyEstimate(mostCommonKey);
return;
}
KeyDetectionResult KeyFinder::findKey(const AudioData& originalAudio, const Parameters& params){
KeyDetectionResult result;
AudioData* workingAudio = new AudioData(originalAudio);
workingAudio->reduceToMono();
// TODO: there is presumably some good maths to determine filter frequencies
float lpfCutoff = params.getLastFrequency() * 1.05;
float dsCutoff = params.getLastFrequency() * 1.10;
unsigned int downsampleFactor = (int)floor( workingAudio->getFrameRate() / 2 / dsCutoff );
// get filter
LowPassFilter* lpf = lpfFactory.getLowPassFilter(160, workingAudio->getFrameRate(), lpfCutoff, 2048);
// feeding in the downsampleFactor for a shortcut
lpf->filter(workingAudio, downsampleFactor);
// note we don't delete the LPF; it's stored in the factory for reuse
Downsampler ds;
ds.downsample(workingAudio, downsampleFactor);
SpectrumAnalyser* sa = new SpectrumAnalyser(workingAudio->getFrameRate(), params, &ctFactory);
// run spectral analysis
Chromagram* ch = sa->chromagram(workingAudio);
delete workingAudio;
delete sa;
// reduce chromagram
ch->reduceTuningBins(params);
result.fullChromagram = Chromagram(*ch);
ch->reduceToOneOctave(params);
result.oneOctaveChromagram = Chromagram(*ch);
// get harmonic change signal
Segmentation* segmenter = Segmentation::getSegmentation(params);
result.harmonicChangeSignal = segmenter->getRateOfChange(*ch, params);
// get track segmentation
std::vector<unsigned int> segmentBoundaries = segmenter->getSegments(result.harmonicChangeSignal, params);
segmentBoundaries.push_back(ch->getHops()); // sentinel
delete segmenter;
// get key estimates for each segment
KeyClassifier hc(params);
std::vector<float> keyWeights(24); // TODO: not ideal using int cast of key_t enum. Hash?
for (int s = 0; s < (signed)segmentBoundaries.size()-1; s++){
KeyDetectionSegment segment;
segment.firstHop = segmentBoundaries[s];
segment.lastHop = segmentBoundaries[s+1] - 1;
// collapse segment's time dimension
std::vector<float> segmentChroma(ch->getBins());
for (unsigned int hop = segment.firstHop; hop <= segment.lastHop; hop++) {
for (unsigned int bin = 0; bin < ch->getBins(); bin++) {
float value = ch->getMagnitude(hop, bin);
segmentChroma[bin] += value;
segment.energy += value;
}
}
segment.key = hc.classify(segmentChroma);
if(segment.key != SILENCE){
keyWeights[segment.key] += segment.energy;
}
result.segments.push_back(segment);
}
delete ch;
// get global key
result.globalKeyEstimate = SILENCE;
float mostCommonKeyWeight = 0.0;
for (int k = 0; k < (signed)keyWeights.size(); k++){
if(keyWeights[k] > mostCommonKeyWeight){
mostCommonKeyWeight = keyWeights[k];
result.globalKeyEstimate = (key_t)k;
}
}
return result;
}