Formant Formant_tracker (Formant me, int ntrack,
double refF1, double refF2, double refF3, double refF4, double refF5,
double dfCost, /* Per kHz. */
double bfCost, double octaveJumpCost)
{
try {
long nformmin = Formant_getMinNumFormants (me);
struct fparm parm;
if (ntrack > nformmin) Melder_throw (U"Number of tracks (", ntrack, U") should not exceed minimum number of formants (", nformmin, U").");
autoFormant thee = Formant_create (my xmin, my xmax, my nx, my dx, my x1, ntrack);
for (long iframe = 1; iframe <= thy nx; iframe ++) {
thy d_frames [iframe]. formant = NUMvector <structFormant_Formant> (1, ntrack);
thy d_frames [iframe]. nFormants = ntrack;
thy d_frames [iframe]. intensity = my d_frames [iframe]. intensity;
}
/* BUG: limit costs to 1e10 or so */
parm.me = me;
parm.thee = thee.peek();
parm.dfCost = dfCost / 1000.0; // per Hz
parm.bfCost = bfCost;
parm.octaveJumpCost = octaveJumpCost;
parm.refF [1] = refF1;
parm.refF [2] = refF2;
parm.refF [3] = refF3;
parm.refF [4] = refF4;
parm.refF [5] = refF5;
NUM_viterbi_multi (my nx, my maxnFormants, ntrack,
getLocalCost, getTransitionCost, putResult, & parm);
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, U": not tracked.");
}
}
Formant Excitation_to_Formant (Excitation me, int maxnFormants)
{
Formant thee = NULL;
long i, nfreq = my nx, nform = 0;
double *p = my z [1];
thee = Formant_create (0, 1, 1, 1, 0.5, maxnFormants); if (! thee) return NULL;
thy frame [1]. formant = NUMstructvector (Formant_Formant, 1, maxnFormants);
for (i = 2; i < nfreq; i ++)
if (p [i] > p [i - 1] && p [i] >= p [i + 1])
{
double min3phon, left, right;
double firstDerivative = p [i+1] - p [i-1], secondDerivative = 2 * p [i] - p [i-1] - p [i+1];
long j;
Formant_Formant formant = & thy frame [1]. formant [++ nform];
formant -> frequency = Excitation_barkToHertz (
my x1 + my dx * (i - 1 + 0.5 * firstDerivative / secondDerivative));
min3phon = p [i] + 0.125 * firstDerivative * firstDerivative / secondDerivative - 3.0;
/* Search left. */
j = i - 1; while (p [j] > min3phon && j > 1) j --;
left = Excitation_barkToHertz (
p [j] > min3phon ? my xmin : my x1 + my dx * (j - 1 + (min3phon - p [j]) / (p [j + 1] - p [j])));
/* Search right. */
j = i + 1; while (p [j] > min3phon && j < nfreq) j ++;
right = Excitation_barkToHertz (
p [j] > min3phon ? my xmax : my x1 + my dx * (j - 1 - (min3phon - p [j]) / (p [j - 1] - p [j])));
formant -> bandwidth = right - left;
if (nform == thy maxnFormants) break;
}
thy frame [1]. nFormants = nform;
return thee;
}
autoFormant Spectrum_to_Formant (Spectrum me, int maxnFormants) {
try {
long nfreq = my nx, nform = 0;
autoNUMvector <double> p (1, nfreq); // power
autoFormant thee = Formant_create (0, 1, 1, 1, 0.5, maxnFormants);
thy d_frames [1]. formant = NUMvector <structFormant_Formant> (1, maxnFormants);
for (long i = 1; i <= nfreq; i ++)
p [i] = my z [1] [i] * my z [1] [i] + my z [2] [i] * my z [2] [i];
for (long i = 2; i < nfreq; i ++)
if (p [i] > p [i - 1] && p [i] >= p [i + 1]) {
double firstDerivative = p [i+1] - p [i-1], secondDerivative = 2 * p [i] - p [i-1] - p [i+1];
Formant_Formant formant = & thy d_frames [1]. formant [++ nform];
formant -> frequency = my dx * (i - 1 + 0.5 * firstDerivative / secondDerivative);
double min3dB = 0.5 * (p [i] + 0.125 * firstDerivative * firstDerivative / secondDerivative);
/* Search left. */
long j = i - 1; while (p [j] > min3dB && j > 1) j --;
if (p [j] > min3dB)
formant -> bandwidth = formant -> frequency;
else
formant -> bandwidth = formant -> frequency - my dx * (j - 1 + (min3dB - p [j]) / (p [j + 1] - p [j]));
/* Search right. */
j = i + 1; while (p [j] > min3dB && j < nfreq) j ++;
if (p [j] > min3dB)
formant -> bandwidth += my xmax - formant -> frequency;
else
formant -> bandwidth += my dx * (j - 1 - (min3dB - p [j]) / (p [j - 1] - p [j])) - formant -> frequency;
if (nform == maxnFormants) break;
}
thy d_frames [1]. nFormants = nform;
return thee;
} catch (MelderError) {
Melder_throw (me, U": not converted to Formant.");
}
}
autoFormant Excitation_to_Formant (Excitation me, int maxnFormants) {
try {
long nfreq = my nx, nform = 0;
double *p = my z [1];
autoFormant thee = Formant_create (0, 1, 1, 1, 0.5, maxnFormants);
thy d_frames [1]. formant = NUMvector <structFormant_Formant> (1, maxnFormants);
for (long i = 2; i < nfreq; i ++)
if (p [i] > p [i - 1] && p [i] >= p [i + 1]) {
double min3phon, left, right;
double firstDerivative = p [i+1] - p [i-1], secondDerivative = 2 * p [i] - p [i-1] - p [i+1];
long j;
Formant_Formant formant = & thy d_frames [1]. formant [++ nform];
formant -> frequency = Excitation_barkToHertz (
my x1 + my dx * (i - 1 + 0.5 * firstDerivative / secondDerivative));
min3phon = p [i] + 0.125 * firstDerivative * firstDerivative / secondDerivative - 3.0;
/* Search left. */
j = i - 1; while (p [j] > min3phon && j > 1) j --;
left = Excitation_barkToHertz (
p [j] > min3phon ? my xmin : my x1 + my dx * (j - 1 + (min3phon - p [j]) / (p [j + 1] - p [j])));
/* Search right. */
j = i + 1; while (p [j] > min3phon && j < nfreq) j ++;
right = Excitation_barkToHertz (
p [j] > min3phon ? my xmax : my x1 + my dx * (j - 1 - (min3phon - p [j]) / (p [j - 1] - p [j])));
formant -> bandwidth = right - left;
if (nform == thy maxnFormants) break;
}
thy d_frames [1]. nFormants = nform;
return thee;
} catch (MelderError) {
Melder_throw (me, U": not converted to Formant.");
}
}
Formant LPC_to_Formant (LPC me, double margin) {
try {
double samplingFrequency = 1.0 / my samplingPeriod;
long nmax = my maxnCoefficients, err = 0;
long interval = nmax > 20 ? 1 : 10;
if (nmax > 99) {
Melder_throw ("We cannot find the roots of a polynomial of order > 99.");
}
if (margin >= samplingFrequency / 4) {
Melder_throw ("Margin must be smaller than ", samplingFrequency / 4, ".");
}
autoFormant thee = Formant_create (my xmin, my xmax, my nx, my dx, my x1, (nmax + 1) / 2);
autoMelderProgress progress (L"LPC to Formant");
for (long i = 1; i <= my nx; i++) {
Formant_Frame formant = & thy d_frames[i];
LPC_Frame lpc = & my d_frames[i];
// Initialisation of Formant_Frame is taken care of in Roots_into_Formant_Frame!
try {
LPC_Frame_into_Formant_Frame (lpc, formant, my samplingPeriod, margin);
} catch (MelderError) {
Melder_clearError();
err++;
}
if ( (interval == 1 || (i % interval) == 1)) {
Melder_progress ( (double) i / my nx, L"LPC to Formant: frame ", Melder_integer (i),
L" out of ", Melder_integer (my nx), L".");
}
}
Formant_sort (thee.peek());
if (err > 0) {
Melder_warning (Melder_integer (err), L" formant frames out of ", Melder_integer (my nx), L" suspect.");
}
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, ": no Formant created.");
}
}
Formant Spectrum_to_Formant (Spectrum me, int maxnFormants)
{
Formant thee = NULL;
long i, nfreq = my nx, nform = 0;
double *p = NUMdvector (1, nfreq); if (! p) goto end; /* power */
thee = Formant_create (0, 1, 1, 1, 0.5, maxnFormants); if (! thee) goto end;
thy frame [1]. formant = NUMstructvector (Formant_Formant, 1, maxnFormants);
if (! thy frame [1]. formant) goto end;
for (i = 1; i <= nfreq; i ++)
p [i] = my z [1] [i] * my z [1] [i] + my z [2] [i] * my z [2] [i];
for (i = 2; i < nfreq; i ++)
if (p [i] > p [i - 1] && p [i] >= p [i + 1])
{
double min3dB;
double firstDerivative = p [i+1] - p [i-1], secondDerivative = 2 * p [i] - p [i-1] - p [i+1];
long j;
Formant_Formant formant = & thy frame [1]. formant [++ nform];
formant -> frequency = my dx * (i - 1 + 0.5 * firstDerivative / secondDerivative);
min3dB = 0.5 * (p [i] + 0.125 * firstDerivative * firstDerivative / secondDerivative);
/* Search left. */
j = i - 1; while (p [j] > min3dB && j > 1) j --;
if (p [j] > min3dB)
formant -> bandwidth = formant -> frequency;
else
formant -> bandwidth = formant -> frequency -
my dx * (j - 1 + (min3dB - p [j]) / (p [j + 1] - p [j]));
/* Search right. */
j = i + 1; while (p [j] > min3dB && j < nfreq) j ++;
if (p [j] > min3dB)
formant -> bandwidth += my xmax - formant -> frequency;
else
formant -> bandwidth += my dx * (j - 1 - (min3dB - p [j]) / (p [j - 1] - p [j])) -
formant -> frequency;
if (nform == maxnFormants) break;
}
thy frame [1]. nFormants = nform;
end:
NUMdvector_free (p, 1);
return thee;
}
Formant FormantGrid_to_Formant (FormantGrid me, double dt, double intensity) {
Melder_assert (dt > 0.0);
Melder_assert (intensity >= 0.0);
long nt = (long) floor ((my xmax - my xmin) / dt) + 1;
double t1 = 0.5 * (my xmin + my xmax - (nt - 1) * dt);
Formant thee = Formant_create (my xmin, my xmax, nt, dt, t1, my formants -> size);
for (long iframe = 1; iframe <= nt; iframe ++) {
Formant_Frame frame = & thy frame [iframe];
frame -> intensity = intensity;
frame -> nFormants = my formants -> size;
frame -> formant = NUMstructvector (Formant_Formant, 1, my formants -> size);
double t = t1 + (iframe - 1) * dt;
for (long iformant = 1; iformant <= my formants -> size; iformant ++) {
Formant_Formant formant = & frame -> formant [iformant];
formant -> frequency = RealTier_getValueAtTime (my formants -> item [iformant], t);
formant -> bandwidth = RealTier_getValueAtTime (my bandwidths -> item [iformant], t);
}
}
end:
iferror forget (thee);
return thee;
}
Formant FormantGrid_to_Formant (FormantGrid me, double dt, double intensity) {
try {
Melder_assert (dt > 0.0);
Melder_assert (intensity >= 0.0);
long nt = (long) floor ((my xmax - my xmin) / dt) + 1;
double t1 = 0.5 * (my xmin + my xmax - (nt - 1) * dt);
autoFormant thee = Formant_create (my xmin, my xmax, nt, dt, t1, my formants -> size);
for (long iframe = 1; iframe <= nt; iframe ++) {
Formant_Frame frame = & thy d_frames [iframe];
frame -> intensity = intensity;
frame -> nFormants = my formants -> size;
frame -> formant = NUMvector <structFormant_Formant> (1, my formants -> size);
double t = t1 + (iframe - 1) * dt;
for (long iformant = 1; iformant <= my formants -> size; iformant ++) {
Formant_Formant formant = & frame -> formant [iformant];
formant -> frequency = RealTier_getValueAtTime ((RealTier) my formants -> item [iformant], t);
formant -> bandwidth = RealTier_getValueAtTime ((RealTier) my bandwidths -> item [iformant], t);
}
}
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, ": not converted to Formant.");
}
}
static autoFormant Sound_to_Formant_any_inline (Sound me, double dt_in, int numberOfPoles,
double halfdt_window, int which, double preemphasisFrequency, double safetyMargin)
{
double dt = dt_in > 0.0 ? dt_in : halfdt_window / 4.0;
double duration = my nx * my dx, t1;
double dt_window = 2.0 * halfdt_window;
long nFrames = 1 + (long) floor ((duration - dt_window) / dt);
long nsamp_window = (long) floor (dt_window / my dx), halfnsamp_window = nsamp_window / 2;
if (nsamp_window < numberOfPoles + 1)
Melder_throw (U"Window too short.");
t1 = my x1 + 0.5 * (duration - my dx - (nFrames - 1) * dt); // centre of first frame
if (nFrames < 1) {
nFrames = 1;
t1 = my x1 + 0.5 * duration;
dt_window = duration;
nsamp_window = my nx;
}
autoFormant thee = Formant_create (my xmin, my xmax, nFrames, dt, t1, (numberOfPoles + 1) / 2); // e.g. 11 poles -> maximally 6 formants
autoNUMvector <double> window (1, nsamp_window);
autoNUMvector <double> frame (1, nsamp_window);
autoNUMvector <double> cof (1, numberOfPoles); // superfluous if which==2, but nobody uses that anyway
autoMelderProgress progress (U"Formant analysis...");
/* Pre-emphasis. */
Sound_preEmphasis (me, preemphasisFrequency);
/* Gaussian window. */
for (long i = 1; i <= nsamp_window; i ++) {
double imid = 0.5 * (nsamp_window + 1), edge = exp (-12.0);
window [i] = (exp (-48.0 * (i - imid) * (i - imid) / (nsamp_window + 1) / (nsamp_window + 1)) - edge) / (1.0 - edge);
}
for (long iframe = 1; iframe <= nFrames; iframe ++) {
double t = Sampled_indexToX (thee.peek(), iframe);
long leftSample = Sampled_xToLowIndex (me, t);
long rightSample = leftSample + 1;
long startSample = rightSample - halfnsamp_window;
long endSample = leftSample + halfnsamp_window;
double maximumIntensity = 0.0;
if (startSample < 1) startSample = 1;
if (endSample > my nx) endSample = my nx;
for (long i = startSample; i <= endSample; i ++) {
double value = Sampled_getValueAtSample (me, i, Sound_LEVEL_MONO, 0);
if (value * value > maximumIntensity) {
maximumIntensity = value * value;
}
}
if (maximumIntensity == HUGE_VAL)
Melder_throw (U"Sound contains infinities.");
thy d_frames [iframe]. intensity = maximumIntensity;
if (maximumIntensity == 0.0) continue; // Burg cannot stand all zeroes
/* Copy a pre-emphasized window to a frame. */
for (long j = 1, i = startSample; j <= nsamp_window; j ++)
frame [j] = Sampled_getValueAtSample (me, i ++, Sound_LEVEL_MONO, 0) * window [j];
if (which == 1) {
burg (frame.peek(), endSample - startSample + 1, cof.peek(), numberOfPoles, & thy d_frames [iframe], 0.5 / my dx, safetyMargin);
} else if (which == 2) {
if (! splitLevinson (frame.peek(), endSample - startSample + 1, numberOfPoles, & thy d_frames [iframe], 0.5 / my dx)) {
Melder_clearError ();
Melder_casual (U"(Sound_to_Formant:)"
U" Analysis results of frame ", iframe,
U" will be wrong."
);
}
}
Melder_progress ((double) iframe / (double) nFrames, U"Formant analysis: frame ", iframe);
}
Formant_sort (thee.peek());
return thee;
}
