bool OME::interpolateResponse(const RealVec &freqs)
{
if(freqs.empty())
{
LOUDNESS_ERROR("OME: Input vector has no elements.");
return 0;
}
else
{
response_.resize(freqs.size(), 0.0);
//load the data into vectors
getData();
//the spline
spline s;
//middle ear
if(middleEarType_ != NONE)
{
s.set_points(middleEarFreqPoints_, middleEardB_);
for (uint i=0; i < freqs.size(); i++)
{
if((freqs[i]<=75) && (middleEarType_ == ANSIS342007_MIDDLE_EAR_HPF))
response_[i] = -14.6; //Post HPF correction
else if(freqs[i] >= middleEarFreqPoints_[40])
response_[i] = middleEardB_[40];
else
response_[i] = s(freqs[i]);
}
}
else
{
LOUDNESS_WARNING("OME: No middle ear filter used.");
}
//outer ear
if(outerEarType_ != NONE)
{
Real lastFreq = outerEarFreqPoints_.back();
Real lastDataPoint = outerEardB_.back();
s.set_points(outerEarFreqPoints_, outerEardB_);
for(uint i = 0; i < freqs.size(); i++)
{
if(freqs[i] >= lastFreq)
response_[i] += lastDataPoint;
else
response_[i] += s(freqs[i]);
}
}
else
{
LOUDNESS_WARNING("OME: No outer ear filter used.");
}
return 1;
}
}
void Model::process(const SignalBank &input)
{
if (initialized_)
modules_[0] -> process(input);
else
LOUDNESS_WARNING(name_ << ": Not initialised!");
}
bool DynamicPartialLoudnessGM::initializeInternal(const TrackBank &input)
{
/*
* Outer-Middle ear filter
*/
//if filter coefficients have not been provided
//use spectral weighting to approximate outer and middle ear response
bool weightSpectrum = false;
if(pathToFilterCoefs_.empty())
{
LOUDNESS_WARNING(name_
<< ": No filter coefficients, opting to weight power spectrum.");
weightSpectrum = true;
//should we use for HPF for low freqs? default is true
if(hpf_)
{
LOUDNESS_DEBUG(name_ << ": Using HPF.");
modules_.push_back(unique_ptr<Module> (new Butter(3, 0, 50.0)));
}
}
else { //otherwise, load them
//load numpy array holding the filter coefficients
cnpy::NpyArray arr = cnpy::npy_load(pathToFilterCoefs_);
Real *data = reinterpret_cast<Real*> (arr.data);
//check if filter is IIR or FIR
bool iir = false;
if(arr.shape[0]==2)
iir = true;
//load the coefficients
RealVec bCoefs, aCoefs;
for(unsigned int i=0; i<arr.shape[1];i++)
{
bCoefs.push_back(data[i]);
if(iir)
aCoefs.push_back(data[i+arr.shape[1]]);
}
//create module
if(iir)
modules_.push_back(unique_ptr<Module>
(new IIR(input.getNTracks(), bCoefs, aCoefs)));
else
modules_.push_back(unique_ptr<Module>
(new FIR(bCoefs)));
//clean up
delete [] data;
}
/*
* Frame generator for spectrogram
*/
int windowSize = round(0.064*input.getFs());
int hopSize = round(timeStep_*input.getFs());
modules_.push_back(unique_ptr<Module>
(new FrameGenerator(windowSize, hopSize)));
/*
* Multi-resolution spectrogram
*/
RealVec bandFreqsHz {10, 80, 500, 1250, 2540, 4050, 15001};
//window spec
RealVec windowSizeSecs {0.064, 0.032, 0.016, 0.008, 0.004, 0.002};
//create appropriate power spectrum module
if(stereoToMono_)
{
modules_.push_back(unique_ptr<Module>
(new PowerSpectrumAndSpatialDetection(bandFreqsHz, windowSizeSecs, uniform_)));
}
else
{
modules_.push_back(unique_ptr<Module>
(new PowerSpectrum(bandFreqsHz, windowSizeSecs, uniform_)));
}
/*
* Compression
*/
if(compressionCriterion_ > 0)
modules_.push_back(unique_ptr<Module>
(new CompressSpectrum(compressionCriterion_)));
/*
* Spectral weighting if necessary
*/
if(weightSpectrum)
{
OME::MiddleEarType middleEar = OME::ANSI;
OME::OuterEarType outerEar = OME::ANSI_FREE;
if(hpf_)
middleEar = OME::ANSI_HPF;
if(diffuseField_)
outerEar = OME::ANSI_DIFFUSE;
modules_.push_back(unique_ptr<Module>
(new WeightSpectrum(middleEar, outerEar)));
}
/*
* Roex filters
*/
if(fastBank_)
{
modules_.push_back(unique_ptr<Module>
(new FastRoexBank(filterSpacing_, interpRoexBank_)));
}
else
{
modules_.push_back(unique_ptr<Module>
(new RoexBankANSIS3407(1.8, 38.9, filterSpacing_)));
}
/*
* Specific loudness
*/
modules_.push_back(unique_ptr<Module>
(new SpecificPartialLoudnessGM()));
/*
* Loudness integration
*/
modules_.push_back(unique_ptr<Module>
(new IntegratedPartialLoudnessGM(diotic_, true)));
return 1;
}
bool PowerSpectrum::initializeInternal(const SignalBank &input)
{
ffts_.clear();
//number of windows
int nWindows = (int)windowSizes_.size();
LOUDNESS_ASSERT(input.getNChannels() == nWindows,
name_ << ": Number of channels do not match number of windows");
LOUDNESS_ASSERT((int)bandFreqsHz_.size() == (nWindows + 1),
name_ << ": Number of frequency bands should equal number of input channels + 1.");
LOUDNESS_ASSERT(!anyAscendingValues(windowSizes_),
name_ << ": Window lengths must be in descending order.");
//work out FFT configuration (constrain to power of 2)
int largestWindowSize = input.getNSamples();
vector<int> fftSize(nWindows, nextPowerOfTwo(largestWindowSize));
if(sampleSpectrumUniformly_)
{
ffts_.push_back(unique_ptr<FFT> (new FFT(fftSize[0])));
ffts_[0] -> initialize();
}
else
{
for(int w=0; w<nWindows; w++)
{
fftSize[w] = nextPowerOfTwo(windowSizes_[w]);
ffts_.push_back(unique_ptr<FFT> (new FFT(fftSize[w])));
ffts_[w] -> initialize();
}
}
//desired bins indices (lo and hi) per band
bandBinIndices_.resize(nWindows);
normFactor_.resize(nWindows);
int fs = input.getFs();
int nBins = 0;
for(int i=0; i<nWindows; i++)
{
//bin indices to use for compiled spectrum
bandBinIndices_[i].resize(2);
//These are NOT the nearest components but satisfies f_k in [f_lo, f_hi)
bandBinIndices_[i][0] = ceil(bandFreqsHz_[i]*fftSize[i]/fs);
// use < bandBinIndices_[i][1] to exclude upper bin
bandBinIndices_[i][1] = ceil(bandFreqsHz_[i+1]*fftSize[i]/fs);
LOUDNESS_ASSERT(bandBinIndices_[i][1]>0,
name_ << ": No components found in band number " << i);
//exclude DC and Nyquist if found
int nyqIdx = (fftSize[i]/2) + (fftSize[i]%2);
if(bandBinIndices_[i][0]==0)
{
LOUDNESS_WARNING(name_ << ": DC found...excluding.");
bandBinIndices_[i][0] = 1;
}
if((bandBinIndices_[i][1]-1) >= nyqIdx)
{
LOUDNESS_WARNING(name_ <<
": Bin is >= nyquist...excluding.");
bandBinIndices_[i][1] = nyqIdx;
}
nBins += bandBinIndices_[i][1]-bandBinIndices_[i][0];
//Power spectrum normalisation
Real refSquared = referenceValue_ * referenceValue_;
switch (normalisation_)
{
case NONE:
normFactor_[i] = 1.0 / refSquared;
break;
case ENERGY:
normFactor_[i] = 2.0/(fftSize[i] * refSquared);
break;
case AVERAGE_POWER:
normFactor_[i] = 2.0/(fftSize[i] * windowSizes_[i] * refSquared);
break;
default:
normFactor_[i] = 2.0/(fftSize[i] * refSquared);
}
LOUDNESS_DEBUG(name_ << ": Normalisation factor : " << normFactor_[i]);
}
//total number of bins in the output spectrum
LOUDNESS_DEBUG(name_
<< ": Total number of bins comprising the output spectrum: " << nBins);
//initialize the output SignalBank
output_.initialize(input.getNEars(), nBins, 1, fs);
output_.setFrameRate(input.getFrameRate());
//output frequencies in Hz
int j = 0, k = 0;
for(int i=0; i<nWindows; i++)
{
j = bandBinIndices_[i][0];
while(j < bandBinIndices_[i][1])
output_.setCentreFreq(k++, (j++)*fs/(Real)fftSize[i]);
LOUDNESS_DEBUG(name_
<< ": Included freq Hz (band low): "
<< fs * bandBinIndices_[i][0] / float(fftSize[i])
<< ": Included freq Hz (band high): "
<< fs * (bandBinIndices_[i][1] - 1) / float(fftSize[i]));
}
return 1;
}
