mrs_realvec Delay::seconds2Samples (mrs_realvec seconds)
{
for (mrs_natural i = 0; i < seconds.getSize (); ++i)
seconds(i) = seconds(i) * israte_;
return seconds;
}
mrs_realvec Delay::samples2Seconds (mrs_realvec samples)
{
for (mrs_natural i = 0; i < samples.getSize (); ++i)
samples(i) = samples(i)/ israte_;
return samples;
}
void TimeFreqPeakConnectivity::InitMatrix (mrs_realvec &Matrix, unsigned char **traceback, mrs_natural rowIdx, mrs_natural colIdx)
{
mrs_natural i,j,
numRows = Matrix.getRows (),
numCols = Matrix.getCols ();
mrs_natural iCol;
Matrix.setval(0);
traceback[rowIdx][colIdx] = kFromLeft;
// left of point of interest
for (i = 0; i < numRows; i++)
{
for (j = 0; j < colIdx; j++)
{
Matrix(i,j) = costInit;
traceback[i][j] = kFromLeft;
}
}
//cout << Matrix << endl;
//upper left corner
for (i = 0; i < rowIdx; i++)
{
iCol = MyMin (rowIdx - i + colIdx, numCols);
for (j = colIdx; j < iCol; j++)
{
Matrix(i,j) = costInit;
traceback[i][j] = kFromLeft;
}
}
//cout << Matrix << endl;
// lower left corner
for (i = rowIdx + 1; i < numRows; i++)
{
iCol = MyMin (i - rowIdx + colIdx, numCols);
for (j = colIdx; j < iCol; j++)
{
Matrix(i,j) = costInit;
traceback[i][j] = kFromLeft;
}
}
//cout << Matrix << endl;
}
void PeakConvert2::ComputePeaker (mrs_realvec in, realvec& out)
{
#ifdef ORIGINAL_VERSION
peaker_->updControl("mrs_real/peakStrength", 0.2);// to be set as a control [!]
#else
peaker_->updControl("mrs_real/peakStrength",1e-1);
peaker_->updControl("mrs_real/peakStrengthRelMax" ,1e-2);
peaker_->updControl("mrs_real/peakStrengthAbs",1e-10 );
peaker_->updControl("mrs_real/peakStrengthTreshLpParam" ,0.95);
peaker_->updControl("mrs_real/peakStrengthRelThresh" , 1.);
#endif
peaker_->updControl("mrs_real/peakSpacing", 2e-3); // 0
peaker_->updControl("mrs_natural/peakStart", downFrequency_); // 0
peaker_->updControl("mrs_natural/peakEnd", upFrequency_); // size_
peaker_->updControl("mrs_natural/inSamples", in.getCols());
peaker_->updControl("mrs_natural/inObservations", in.getRows());
peaker_->updControl("mrs_natural/onSamples", out.getCols());
peaker_->updControl("mrs_natural/onObservations", out.getRows());
peaker_->process(in, out);
}
void TimeFreqPeakConnectivity::CalcDp (mrs_realvec &Matrix, mrs_natural startr, mrs_natural startc, mrs_natural stopr, mrs_natural stopc)
{
mrs_natural i,j,
numRows = Matrix.getRows (),
numCols = Matrix.getCols ();
mrs_real prevCost[kNumDirections] = {0,0,0};
costMatrix_.stretch ( numRows, numCols);
// initialize cost and traceback matrix
// upper and lower left corner
InitMatrix (costMatrix_, tracebackIdx_, startr, startc);
costMatrix_(startr, startc) = Matrix(startr, startc);
// compute cost matrix
for (j = startc+1; j <= stopc; j++)
{
mrs_natural rowLoopStart = MyMax (0, startr - (j - startc)),
rowLoopStop = MyMin (numRows, startr + (j-startc) + 1);
for (i = rowLoopStart; i < rowLoopStop; i++)
{
prevCost[kFromLeft] = costMatrix_(i,j-1);
prevCost[kFromUp] = (i >= numRows-1) ? costInit : costMatrix_(i+1, j-1); // the if isn't very nice here....
prevCost[kFromDown] = (i <= 0) ? costInit : costMatrix_(i-1, j-1);
// assign cost
costMatrix_(i,j) = Matrix(i,j) + dtwFindMin (prevCost, tracebackIdx_[i][j]);
}
}
// compute path
i = stopr;
for (j = stopc; j >= startc; j--)
{
path_[j-startc] = i;
i -= (kFromLeft - tracebackIdx_[i][j]); // note: does work only for kFromUp, kFromLeft, kFromDown
}
}
static void FreqSmear (mrs_realvec &spectrum)
{
mrs_natural length = spectrum.getSize ();
mrs_real buf[3] = {0,0,0},
coeff[3] = {.25, .5, .25};
spectrum(0) = spectrum(length-1) = 0;
for (mrs_natural i = 0; i < length-1; i++)
{
buf[(i+1)%3] = spectrum(i+1);
spectrum(i) = coeff[0]*buf[(i-1+3)%3] + coeff[1]*buf[(i)%3] + coeff[2]*buf[(i+1)%3];
}
return;
}
void Peaker::compLpThresh (const mrs_realvec input, mrs_realvec &output)
{
mrs_natural i,
len = input.getCols ();
mrs_real buf = input(0);
for (i = 0; i < len; ++i)
{
buf = input(i) + lpCoeff_ * (buf - input(i));
output(i) = buf;
}
// do reverse filtering to ensure zero group delay
for (i = len-1; i >= 0; i--)
{
buf = output(i) + lpCoeff_ * (buf - output(i));
output(i) = buf;
}
}
// used inside myProcess
mrs_real
HarmonicStrength::quadratic_interpolation(mrs_real best_bin,
mrs_realvec& in, mrs_natural t)
{
if ((best_bin == 0) || (best_bin == in.getRows()-1))
{
// don't try to interpolate using data that's
// outside of the spectogram
// TODO: find some convincing DSP thing to do in this case
return in( (mrs_natural)best_bin, t);
}
// https://ccrma.stanford.edu/~jos/sasp/Quadratic_Interpolation_Spectral_Peaks.html
// a = alpha, b = beta, g = gamma
mrs_real a = in( (mrs_natural)best_bin - 1, t);
mrs_real b = in( (mrs_natural)best_bin + 0, t);
mrs_real g = in( (mrs_natural)best_bin + 1, t);
mrs_real p = 0.5 * (a-g)/(a-2*b+g);
// avoid some NaNs
if ((p < -0.5) || (p > 0.5))
{
return b;
}
mrs_real yp = b - 0.25*(a-g)*p;
// avoid all NaNs
if (yp < b)
{
// I think this happens because the search window doesn't
// encompass the entire spectrum, so the "highest" bin
// might not actually be the highest one, if it was on the
// edge of search window.
// TODO: find some convincing DSP thing to do in this case
return b;
}
return yp;
}
void
PeakDistanceHorizontality::myProcess(realvec& in, realvec& out)
{
mrs_natural i;
const mrs_natural numInputs = getctrl ("mrs_natural/numInputs")->to<mrs_natural>();
const mrs_realvec isHoriz = ctrl_horizvert_->to<mrs_realvec>();
const mrs_real range[2] = {ctrl_rangeX_->to<mrs_real>(), ctrl_rangeY_->to<mrs_real>()};
out = in;
MRSASSERT(range[0] > 0 && range[1] > 0);
if (isHoriz.getSize () != numInputs)
{
MRSWARN("PeakDistanceHorizontality: dimension mismatch");
MRSASSERT(false);
out.setval(0);
return;
}
if (getctrl("mrs_bool/bypass")->to<mrs_bool>())
{
weights_.setval(1.);
setctrl ("mrs_realvec/weights", weights_);
return;
}
for (i = 0; i < inSamples_; i++)
{
for (mrs_natural j = i; j < inSamples_; j++)
{
mrs_natural k;
mrs_real horizontality = ComputeHorizontality ( std::abs(in(1,i)-in(1,j))/range[0],
std::abs(in(0,i)-in(0,j))/range[1]),
norm = 0;
for (k = 0; k < numInputs; k++)
{
mrs_real weight = horizontality;
if (abs(isHoriz(k) - 2) < kIdentityThresh)
weight = .5; // input is both horizontal and vertical
else if (abs(isHoriz(k)) < kIdentityThresh)
weight = 1.-weight; // input is vertical
norm += weight;
weights_(k*inSamples_ + i, j) = weight;
weights_(k*inSamples_ + j, i) = weight; // symmetry
}
if (norm != 0)
norm = 1./norm;
for (k = 0; k < numInputs; k++)
{
weights_(k*inSamples_ + i, j) *= norm;
if (i != j)
weights_(k*inSamples_ + j, i) *= norm; // symmetry
}
}
}
setctrl ("mrs_realvec/weights", weights_);
#ifdef MARSYAS_MATLAB
#ifdef MTLB_DBG_LOG
MATLAB_PUT(weights_, "weights");
MATLAB_EVAL("figure(1);imagesc(weights),colorbar;");
#endif
#endif
}
void
SimulMaskingFft::CalcSpreading (mrs_realvec &bandLevels, mrs_realvec &result)
{
// this is level dependent adapted from ITU-R BS.1387
mrs_natural iBarkj, // Masker
iBarkk; // Maskee
mrs_real fTmp1,
fTmp2,
fSlope,
fScale = sqrt(8./3.),
fBRes = barkRes_,//hertz2bark (.5*audiosrate_, h2bIdx)/numBands_,
*pfEnPowTmp = processBuff_.getData (),
*pfSlopeUp = helpBuff_.getData ();
mrs_real *pfSlope = slopeSpread_.getData (),
*pfNorm = normSpread_.getData ();
// initialize pfResult
result.setval(0);
fSlope = exp ( -fBRes * 2.7 * 2.302585092994045684017991454684364207601101488628772976033);
fTmp2 = 1.0 / (1.0 - fSlope);
for (iBarkj = 0; iBarkj < numBands_; iBarkj++)
{
pfSlopeUp[iBarkj] = pfSlope[iBarkj] * pow (bandLevels(iBarkj)*fScale, .2 * fBRes);
fTmp1 = (1.0 - IntPow (fSlope, iBarkj+1)) * fTmp2;
fTmp2 = (1.0 - IntPow(pfSlopeUp[iBarkj], numBands_ - iBarkj)) / (1.0 - pfSlopeUp[iBarkj]);
if (bandLevels(iBarkj) < 1e-20)
{
pfSlopeUp[iBarkj] = 0;
pfEnPowTmp[iBarkj] = 0;
continue;
}
pfSlopeUp[iBarkj] = exp (0.4 * log (pfSlopeUp[iBarkj]));
pfEnPowTmp[iBarkj] = exp (0.4 * log (bandLevels(iBarkj)/(fTmp1 + fTmp2 -1)));
}
fSlope = exp ( 0.4 * log (fSlope));
// lower slope
result(numBands_-1) = pfEnPowTmp[numBands_-1];
for (iBarkk = numBands_-2; iBarkk >= 0; iBarkk--)
result(iBarkk) = pfEnPowTmp[iBarkk] + result(iBarkk + 1) * fSlope;
// upper slope
for (iBarkj = 0; iBarkj < numBands_-1; iBarkj++)
{
fSlope = pfSlopeUp[iBarkj];
fTmp1 = pfEnPowTmp[iBarkj];
for (iBarkk = iBarkj+1; iBarkk < numBands_; iBarkk++)
{
mrs_real dTmp1 = fTmp1 * fSlope;
fTmp1 = (dTmp1 < 1e-30)? 0 : (mrs_real)dTmp1;
result(iBarkk) += fTmp1;
}
}
// normalization
for (iBarkk = 0; iBarkk < numBands_; iBarkk++)
{
mrs_real dTmp = result(iBarkk);
result(iBarkk) = sqrt(dTmp) * dTmp * dTmp *pfNorm[iBarkk];
//result(iBarkk) = sqrt (result(iBarkk));
//result(iBarkk) *= result(iBarkk)*result(iBarkk)*result(iBarkk)*result(iBarkk);
//result(iBarkk) *= pfNorm[iBarkk];
}
return;
}
