Matrix Spectrogram_to_Matrix (Spectrogram me) {
try {
autoMatrix thee = Matrix_create (my xmin, my xmax, my nx, my dx, my x1, my ymin, my ymax, my ny, my dy, my y1);
NUMmatrix_copyElements (my z, thy z, 1, my ny, 1, my nx);
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, U": not converted to Matrix.");
}
}
autoActivation Matrix_to_Activation (Matrix me) {
try {
autoActivation thee = Activation_create (my ny, my nx);
NUMmatrix_copyElements (my z, thy z, 1, my ny, 1, my nx);
return thee;
} catch (MelderError) {
Melder_throw (me, U": not converted to Activation.");
}
}
autoMatrix Activation_to_Matrix (Activation me) {
try {
autoMatrix thee = Matrix_create (my xmin, my xmax, my nx, my dx, my x1, my ymin, my ymax, my ny, my dy, my y1);
NUMmatrix_copyElements (my z, thy z, 1, my ny, 1, my nx);
return thee;
} catch (MelderError) {
Melder_throw (me, U": not converted to Matrix.");
}
}
autoMelSpectrogram Matrix_to_MelSpectrogram (Matrix me) {
try {
autoMelSpectrogram thee = MelSpectrogram_create (my xmin, my xmax, my nx, my dx, my x1, my ymin, my ymax, my ny, my dy, my y1);
NUMmatrix_copyElements (my z, thy z, 1, my ny, 1, my nx);
return thee;
} catch (MelderError) {
Melder_throw (me, U": not converted to MelSpectrogram.");
}
}
autoTableOfReal SVD_extractRightSingularVectors (SVD me) {
try {
long mn_min = MIN (my numberOfRows, my numberOfColumns);
autoTableOfReal thee = TableOfReal_create (my numberOfColumns, mn_min);
NUMmatrix_copyElements (my v, thy data, 1, my numberOfColumns, 1, mn_min);
return thee;
} catch (MelderError) {
Melder_throw (me, U": right singular vector not extracted.");
}
}
TableOfReal SVD_extractLeftSingularVectors (I) {
iam (SVD);
try {
long mn_min = MIN (my numberOfRows, my numberOfColumns);
autoTableOfReal thee = TableOfReal_create (my numberOfRows, mn_min);
NUMmatrix_copyElements (my u, thy data, 1, my numberOfRows, 1, mn_min);
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, ": left singular vector not extracted.");
}
}
autoConfiguration TableOfReal_to_Configuration (TableOfReal me) {
try {
autoConfiguration thee = Configuration_create (my numberOfRows, my numberOfColumns);
NUMmatrix_copyElements (my data, thy data, 1, my numberOfRows, 1, my numberOfColumns);
TableOfReal_copyLabels (me, thee.peek(), 1, 1);
return thee;
} catch (MelderError) {
Melder_throw (me, U": not converted.");
}
}
FormantFilter Matrix_to_FormantFilter (I) {
iam (Matrix);
try {
autoFormantFilter thee = FormantFilter_create (my xmin, my xmax, my nx, my dx, my x1,
my ymin, my ymax, my ny, my dy, my y1);
NUMmatrix_copyElements (my z, thy z, 1, my ny, 1, my nx);
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, U": not converted to FormantFilter.");
}
}
autoAffineTransform Configurations_to_AffineTransform_congruence (Configuration me, Configuration thee, long maximumNumberOfIterations, double tolerance) {
try {
// Use Procrustes transform to obtain starting configuration.
// (We only need the transformation matrix T.)
autoProcrustes p = Configurations_to_Procrustes (me, thee, 0);
NUMmaximizeCongruence (my data, thy data, my numberOfRows, p -> n, p -> r, maximumNumberOfIterations, tolerance);
autoAffineTransform at = AffineTransform_create (p -> n);
NUMmatrix_copyElements (p -> r, at -> r, 1, p -> n, 1, p -> n);
return at;
} catch (MelderError) {
Melder_throw (me, U": no congruence transformation created.");
}
}
TableOfReal AffineTransform_extractMatrix (I) {
iam (AffineTransform);
try {
autoTableOfReal thee = TableOfReal_create (my n, my n);
NUMmatrix_copyElements (my r, thy data, 1, my n, 1, my n);
for (long i = 1; i <= my n; i++) {
wchar_t label[20];
(void) swprintf (label, 20, L"%ld", i);
TableOfReal_setRowLabel (thee.peek(), i, label);
TableOfReal_setColumnLabel (thee.peek(), i, label);
}
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, ": transformation matrix not extracted.");
}
}
static void Diagonalizer_and_CrossCorrelationTable_qdiag (Diagonalizer me, CrossCorrelationTables thee, double *cweights, long maxNumberOfIterations, double delta) {
try {
CrossCorrelationTable c0 = (CrossCorrelationTable) thy item[1];
double **w = my data;
long dimension = c0 -> numberOfColumns;
autoEigen eigen = Thing_new (Eigen);
autoCrossCorrelationTables ccts = Data_copy (thee);
autoNUMmatrix<double> pinv (1, dimension, 1, dimension);
autoNUMmatrix<double> d (1, dimension, 1, dimension);
autoNUMmatrix<double> p (1, dimension, 1, dimension);
autoNUMmatrix<double> m1 (1, dimension, 1, dimension);
autoNUMmatrix<double> wc (1, dimension, 1, dimension);
autoNUMvector<double> wvec (1, dimension);
autoNUMvector<double> wnew (1, dimension);
autoNUMvector<double> mvec (1, dimension);
for (long i = 1; i <= dimension; i++) // Transpose W
for (long j = 1; j <= dimension; j++) {
wc[i][j] = w[j][i];
}
// d = diag(diag(W'*C0*W));
// W = W*d^(-1/2);
NUMdmatrix_normalizeColumnVectors (wc.peek(), dimension, dimension, c0 -> data);
// scale eigenvectors for sphering
// [vb,db] = eig(C0);
// P = db^(-1/2)*vb';
Eigen_initFromSymmetricMatrix (eigen.peek(), c0 -> data, dimension);
for (long i = 1; i <= dimension; i++) {
if (eigen -> eigenvalues[i] < 0) {
Melder_throw (U"Covariance matrix not positive definite, eigenvalue[", i, U"] is negative.");
}
double scalef = 1 / sqrt (eigen -> eigenvalues[i]);
for (long j = 1; j <= dimension; j++) {
p[dimension - i + 1][j] = scalef * eigen -> eigenvectors[i][j];
}
}
// P*C[i]*P'
for (long ic = 1; ic <= thy size; ic++) {
CrossCorrelationTable cov1 = (CrossCorrelationTable) thy item[ic];
CrossCorrelationTable cov2 = (CrossCorrelationTable) ccts -> item[ic];
NUMdmatrices_multiply_VCVp (cov2 -> data, p.peek(), dimension, dimension, cov1 -> data, 1);
}
// W = P'\W == inv(P') * W
NUMpseudoInverse (p.peek(), dimension, dimension, pinv.peek(), 0);
NUMdmatrices_multiply_VpC (w, pinv.peek(), dimension, dimension, wc.peek(), dimension);
// initialisation for order KN^3
for (long ic = 2; ic <= thy size; ic++) {
CrossCorrelationTable cov = (CrossCorrelationTable) ccts -> item[ic];
// C * W
NUMdmatrices_multiply_VC (m1.peek(), cov -> data, dimension, dimension, w, dimension);
// D += scalef * M1*M1'
NUMdmatrices_multiplyScaleAdd (d.peek(), m1.peek(), dimension, dimension, 2 * cweights[ic]);
}
long iter = 0;
double delta_w;
autoMelderProgress progress (U"Simultaneous diagonalization of many CrossCorrelationTables...");
try {
do {
// the standard diagonality measure is rather expensive to calculate so we compare the norms of
// differences of eigenvectors.
delta_w = 0;
for (long kol = 1; kol <= dimension; kol++) {
for (long i = 1; i <= dimension; i++) {
wvec[i] = w[i][kol];
}
update_one_column (ccts.peek(), d.peek(), cweights, wvec.peek(), -1, mvec.peek());
Eigen_initFromSymmetricMatrix (eigen.peek(), d.peek(), dimension);
// Eigenvalues already sorted; get eigenvector of smallest !
for (long i = 1; i <= dimension; i++) {
wnew[i] = eigen -> eigenvectors[dimension][i];
}
update_one_column (ccts.peek(), d.peek(), cweights, wnew.peek(), 1, mvec.peek());
for (long i = 1; i <= dimension; i++) {
w[i][kol] = wnew[i];
}
// compare norms of eigenvectors. We have to compare ||wvec +/- w_new|| because eigenvectors
// may change sign.
double normp = 0, normm = 0;
for (long j = 1; j <= dimension; j++) {
double dm = wvec[j] - wnew[j], dp = wvec[j] + wnew[j];
normp += dm * dm; normm += dp * dp;
}
normp = normp < normm ? normp : normm;
normp = sqrt (normp);
delta_w = normp > delta_w ? normp : delta_w;
}
iter++;
Melder_progress ((double) iter / (double) (maxNumberOfIterations + 1), U"Iteration: ", iter, U", norm: ", delta_w);
} while (delta_w > delta && iter < maxNumberOfIterations);
} catch (MelderError) {
Melder_clearError ();
}
// Revert the sphering W = P'*W;
// Take transpose to make W*C[i]W' diagonal instead of W'*C[i]*W => (P'*W)'=W'*P
NUMmatrix_copyElements (w, wc.peek(), 1, dimension, 1, dimension);
NUMdmatrices_multiply_VpC (w, wc.peek(), dimension, dimension, p.peek(), dimension); // W = W'*P: final result
// Calculate the "real" diagonality measure
// double dm = CrossCorrelationTables_and_Diagonalizer_getDiagonalityMeasure (thee, me, cweights, 1, thy size);
} catch (MelderError) {
Melder_throw (me, U" & ", thee, U": no joint diagonalization (qdiag).");
}
}
/*
This routine is modeled after qdiag.m from Andreas Ziehe, Pavel Laskov, Guido Nolte, Klaus-Robert Müller,
A Fast Algorithm for Joint Diagonalization with Non-orthogonal Transformations and its Application to
Blind Source Separation, Journal of Machine Learning Research 5 (2004), 777–800.
*/
static void Diagonalizer_and_CrossCorrelationTables_ffdiag (Diagonalizer me, CrossCorrelationTables thee, long maxNumberOfIterations, double delta) {
try {
long iter = 0, dimension = my numberOfRows;
double **v = my data;
autoCrossCorrelationTables ccts = CrossCorrelationTables_and_Diagonalizer_diagonalize (thee, me);
autoNUMmatrix<double> w (1, dimension, 1, dimension);
autoNUMmatrix<double> vnew (1, dimension, 1, dimension);
autoNUMmatrix<double> cc (1, dimension, 1, dimension);
for (long i = 1; i <= dimension; i++) {
w[i][i] = 1;
}
autoMelderProgress progress (U"Simultaneous diagonalization of many CrossCorrelationTables...");
double dm_new = CrossCorrelationTables_getDiagonalityMeasure (ccts.peek(), nullptr, 0, 0);
try {
double dm_old, theta = 1, dm_start = dm_new;
do {
dm_old = dm_new;
for (long i = 1; i <= dimension; i++) {
for (long j = i + 1; j <= dimension; j++) {
double zii = 0, zij = 0, zjj = 0, yij = 0, yji = 0; // zij = zji
for (long k = 1; k <= ccts -> size; k++) {
CrossCorrelationTable ct = (CrossCorrelationTable) ccts -> item [k];
zii += ct -> data[i][i] * ct -> data[i][i];
zij += ct -> data[i][i] * ct -> data[j][j];
zjj += ct -> data[j][j] * ct -> data[j][j];
yij += ct -> data[j][j] * ct -> data[i][j];
yji += ct -> data[i][i] * ct -> data[i][j];
}
double denom = zjj * zii - zij * zij;
if (denom != 0) {
w[i][j] = (zij * yji - zii * yij) / denom;
w[j][i] = (zij * yij - zjj * yji) / denom;
}
}
}
double norma = 0;
for (long i = 1; i <= dimension; i++) {
double normai = 0;
for (long j = 1; j <= dimension; j++) {
if (i != j) {
normai += fabs (w[i][j]);
}
}
if (normai > norma) {
norma = normai;
}
}
// evaluate the norm
if (norma > theta) {
double normf = 0;
for (long i = 1; i <= dimension; i++)
for (long j = 1; j <= dimension; j++)
if (i != j) {
normf += w[i][j] * w[i][j];
}
double scalef = theta / sqrt (normf);
for (long i = 1; i <= dimension; i++) {
for (long j = 1; j <= dimension; j++) {
if (i != j) {
w[i][j] *= scalef;
}
}
}
}
// update V
NUMmatrix_copyElements (v, vnew.peek(), 1, dimension, 1, dimension);
NUMdmatrices_multiply_VC (v, w.peek(), dimension, dimension, vnew.peek(), dimension);
for (long k = 1; k <= ccts -> size; k++) {
CrossCorrelationTable ct = (CrossCorrelationTable) ccts -> item[k];
NUMmatrix_copyElements (ct -> data, cc.peek(), 1, dimension, 1, dimension);
NUMdmatrices_multiply_VCVp (ct -> data, w.peek(), dimension, dimension, cc.peek(), 1);
}
dm_new = CrossCorrelationTables_getDiagonalityMeasure (ccts.peek(), 0, 0, 0);
iter++;
Melder_progress ((double) iter / (double) maxNumberOfIterations, U"Iteration: ", iter, U", measure: ", dm_new, U"\n fractional measure: ", dm_new / dm_start);
} while (fabs ((dm_old - dm_new) / dm_new) > delta && iter < maxNumberOfIterations);
} catch (MelderError) {
Melder_clearError ();
}
} catch (MelderError) {
Melder_throw (me, U" & ", thee, U": no joint diagonalization (ffdiag).");
}
}
Sound Sound_deepenBandModulation (Sound me, double enhancement_dB,
double flow, double fhigh, double slowModulation, double fastModulation, double bandSmoothing)
{
try {
autoSound thee = Data_copy (me);
double maximumFactor = pow (10, enhancement_dB / 20), alpha = sqrt (log (2.0));
double alphaslow = alpha / slowModulation, alphafast = alpha / fastModulation;
for (long channel = 1; channel <= my ny; channel ++) {
autoSound channelSound = Sound_extractChannel (me, channel);
autoSpectrum orgspec = Sound_to_Spectrum (channelSound.peek(), true);
/*
* Keep the part of the sound that is outside the filter bank.
*/
autoSpectrum spec = Data_copy (orgspec.peek());
Spectrum_stopHannBand (spec.peek(), flow, fhigh, bandSmoothing);
autoSound filtered = Spectrum_to_Sound (spec.peek());
long n = thy nx;
double *amp = thy z [channel];
for (long i = 1; i <= n; i ++) amp [i] = filtered -> z [1] [i];
autoMelderProgress progress (U"Deepen band modulation...");
double fmin = flow;
while (fmin < fhigh) {
/*
* Take a one-bark frequency band.
*/
double fmid_bark = NUMhertzToBark (fmin) + 0.5, ceiling;
double fmax = NUMbarkToHertz (NUMhertzToBark (fmin) + 1);
if (fmax > fhigh) fmax = fhigh;
Melder_progress (fmin / fhigh, U"Band: ", Melder_fixed (fmin, 0), U" ... ", Melder_fixed (fmax, 0), U" Hz");
NUMmatrix_copyElements (orgspec -> z, spec -> z, 1, 2, 1, spec -> nx);
Spectrum_passHannBand (spec.peek(), fmin, fmax, bandSmoothing);
autoSound band = Spectrum_to_Sound (spec.peek());
/*
* Compute a relative intensity contour.
*/
autoSound intensity = Data_copy (band.peek());
n = intensity -> nx;
amp = intensity -> z [1];
for (long i = 1; i <= n; i ++) amp [i] = 10 * log10 (amp [i] * amp [i] + 1e-6);
autoSpectrum intensityFilter = Sound_to_Spectrum (intensity.peek(), true);
n = intensityFilter -> nx;
for (long i = 1; i <= n; i ++) {
double frequency = intensityFilter -> x1 + (i - 1) * intensityFilter -> dx;
double slow = alphaslow * frequency, fast = alphafast * frequency;
double factor = exp (- fast * fast) - exp (- slow * slow);
intensityFilter -> z [1] [i] *= factor;
intensityFilter -> z [2] [i] *= factor;
}
intensity.reset (Spectrum_to_Sound (intensityFilter.peek()));
n = intensity -> nx;
amp = intensity -> z [1];
for (long i = 1; i <= n; i ++) amp [i] = pow (10, amp [i] / 2);
/*
* Clip to maximum enhancement.
*/
ceiling = 1 + (maximumFactor - 1.0) * (0.5 - 0.5 * cos (NUMpi * fmid_bark / 13));
for (long i = 1; i <= n; i ++) amp [i] = 1 / (1 / amp [i] + 1 / ceiling);
n = thy nx;
amp = thy z [channel];
for (long i = 1; i <= n; i ++) amp [i] += band -> z [1] [i] * intensity -> z [1] [i];
fmin = fmax;
}
}
Vector_scale (thee.peek(), 0.99);
/* Truncate. */
thy xmin = my xmin;
thy xmax = my xmax;
thy nx = my nx;
thy x1 = my x1;
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, U": band modulation not deepened.");
}
}
/*
Varimax rotation, implementation according to:
Jos Ten Berge (1995), "Suppressing permutations or rigid
planar rotations: a remedy against nonoptimal varimax rotations",
Psychometrika 60, 437-446.
*/
static void NUMvarimax (double **xm, double **ym, long nr, long nc, int normalizeRows, int quartimax, long maximumNumberOfIterations, double tolerance) {
Melder_assert (nr > 0 && nc > 0);
NUMmatrix_copyElements (xm, ym, 1, nr, 1, nc);
if (nc == 1) {
return;
}
if (nc == 2) {
maximumNumberOfIterations = 1;
}
autoNUMvector<double> u (1, nr);
autoNUMvector<double> v (1, nr);
autoNUMvector<double> norm;
// Normalize sum of squares of each row to one.
// After rotation we have to rescale.
if (normalizeRows) {
norm.reset (1, nr);
for (long i = 1; i <= nr; i++) {
for (long j = 1; j <= nc; j++) {
norm[i] += ym[i][j] * ym[i][j];
}
if (norm[i] <= 0.0) {
continue;
}
norm[i] = sqrt (norm[i]);
for (long j = 1; j <= nc; j++) {
ym[i][j] /= norm[i];
}
}
}
// Initial squared "variance".
double varianceSq = NUMsquaredVariance (ym, nr, nc, quartimax);
if (varianceSq == 0.0) {
return;
}
// Treat columns pairwise.
double varianceSq_old;
long numberOfIterations = 0;
do {
for (long c1 = 1; c1 <= nc; c1++) {
for (long c2 = c1 + 1; c2 <= nc; c2++) {
double um = 0.0, vm = 0.0;
for (long i = 1; i <= nr; i++) {
double x = ym[i][c1], y = ym[i][c2];
u[i] = x * x - y * y;
um += u[i];
v[i] = 2.0 * x * y;
vm += v[i];
}
um /= nr; vm /= nr;
if (quartimax || nr == 1) {
um = vm = 0.0;
}
/*
In the paper just before equation (1):
a = 2n u'v, b = n(u'u-v'v), c = sqrt(a^2+b^2)
w = -sign(a) sqrt((b+c) / 2c)
Tricks: multiplication with n drops out!
a's multiplication by 2 outside the loop.
*/
double a = 0.0, b = 0.0;
for (long i = 1; i <= nr; i++) {
double ui = u[i] - um, vi = v[i] - vm;
a += ui * vi;
b += ui * ui - vi * vi;
}
double c = sqrt (4.0 * a * a + b * b);
double w = sqrt ( (c + b) / (2.0 * c));
if (a > 0.0) {
w = -w;
}
double cost = sqrt (0.5 + 0.5 * w);
double sint = sqrt (0.5 - 0.5 * w);
double t22 = cost;
double t11 = cost;
double t12 = sint;
double t21 = -sint;
// Prevent permutations: when w < 0, i.e., a > 0, swap columns of T:/
if (w < 0.0) {
t11 = sint; t12 = t21 = cost; t22 = -sint;
}
// Rotate in the plane spanned by c1 and c2.
for (long i = 1; i <= nr; i++) {
double *xt = ym[i], xtc1 = xt[c1];
xt[c1] = xtc1 * t11 + xt[c2] * t21;
xt[c2] = xtc1 * t12 + xt[c2] * t22;
}
}
}
numberOfIterations++;
varianceSq_old = varianceSq;
varianceSq = NUMsquaredVariance (ym, nr, nc, quartimax);
} while (fabs (varianceSq_old - varianceSq) / varianceSq_old > tolerance &&
numberOfIterations < maximumNumberOfIterations);
if (normalizeRows) {
for (long i = 1; i <= nr; i++) {
for (long j = 1; j <= nc; j++) {
ym[i][j] *= norm[i];
}
}
}
}
