Sound EEG_to_Sound_modulated (EEG me, double baseFrequency, double channelBandwidth, const wchar_t *channelRanges) {
try {
long numberOfChannels;
autoNUMvector <long> channelNumbers (NUMstring_getElementsOfRanges (channelRanges, my d_numberOfChannels, & numberOfChannels, NULL, L"channel", true), 1);
double maxFreq = baseFrequency + my d_numberOfChannels * channelBandwidth;
double samplingFrequency = 2 * maxFreq;
samplingFrequency = samplingFrequency < 44100 ? 44100 : samplingFrequency;
autoSound thee = Sound_createSimple (1, my xmax - my xmin, samplingFrequency);
for (long i = 1; i <= numberOfChannels; i++) {
long ichannel = channelNumbers[i];
double fbase = baseFrequency;// + (ichannel - 1) * channelBandwidth;
autoSound si = Sound_extractChannel (my d_sound, ichannel);
autoSpectrum spi = Sound_to_Spectrum (si.peek(), 1);
Spectrum_passHannBand (spi.peek(), 0.5, channelBandwidth - 0.5, 0.5);
autoSpectrum spi_shifted = Spectrum_shiftFrequencies (spi.peek(), fbase, samplingFrequency / 2, 30);
autoSound resampled = Spectrum_to_Sound (spi_shifted.peek());
long nx = resampled -> nx < thy nx ? resampled -> nx : thy nx;
for (long j = 1; j <= nx; j++) {
thy z[1][j] += resampled -> z[1][j];
}
}
Vector_scale (thee.peek(), 0.99);
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, ": no playable sound created.");
}
}
autoSound Spectrum_to_Sound (Spectrum me) {
try {
double *re = my z [1], *im = my z [2];
double lastFrequency = my x1 + (my nx - 1) * my dx;
int originalNumberOfSamplesProbablyOdd = im [my nx] != 0.0 || my xmax - lastFrequency > 0.25 * my dx;
if (my x1 != 0.0)
Melder_throw (U"A Fourier-transformable Spectrum must have a first frequency of 0 Hz, not ", my x1, U" Hz.");
long numberOfSamples = 2 * my nx - ( originalNumberOfSamplesProbablyOdd ? 1 : 2 );
autoSound thee = Sound_createSimple (1, 1 / my dx, numberOfSamples * my dx);
double *amp = thy z [1];
double scaling = my dx;
amp [1] = re [1] * scaling;
for (long i = 2; i < my nx; i ++) {
amp [i + i - 1] = re [i] * scaling;
amp [i + i] = im [i] * scaling;
}
if (originalNumberOfSamplesProbablyOdd) {
amp [numberOfSamples] = re [my nx] * scaling;
if (numberOfSamples > 1) amp [2] = im [my nx] * scaling;
} else {
amp [2] = re [my nx] * scaling;
}
NUMrealft (amp, numberOfSamples, -1);
return thee;
} catch (MelderError) {
Melder_throw (me, U": not converted to Sound.");
}
}
static LPC _Sound_to_LPC (Sound me, int predictionOrder, double analysisWidth, double dt,
double preEmphasisFrequency, int method, double tol1, double tol2) {
double t1, samplingFrequency = 1.0 / my dx;
double windowDuration = 2 * analysisWidth; /* gaussian window */
long nFrames, frameErrorCount = 0;
if (floor (windowDuration / my dx) < predictionOrder + 1) Melder_throw ("Analysis window duration too short.\n"
"For a prediction order of ", predictionOrder, " the analysis window duration has to be greater than ", my dx * (predictionOrder + 1),
"Please increase the analysis window duration or lower the prediction order.");
// Convenience: analyse the whole sound into one LPC_frame
if (windowDuration > my dx * my nx) {
windowDuration = my dx * my nx;
}
Sampled_shortTermAnalysis (me, windowDuration, dt, & nFrames, & t1);
autoSound sound = Data_copy (me);
autoSound sframe = Sound_createSimple (1, windowDuration, samplingFrequency);
autoSound window = Sound_createGaussian (windowDuration, samplingFrequency);
autoLPC thee = LPC_create (my xmin, my xmax, nFrames, dt, t1, predictionOrder, my dx);
autoMelderProgress progress (L"LPC analysis");
if (preEmphasisFrequency < samplingFrequency / 2) {
Sound_preEmphasis (sound.peek(), preEmphasisFrequency);
}
for (long i = 1; i <= nFrames; i++) {
LPC_Frame lpcframe = (LPC_Frame) & thy d_frames[i];
double t = Sampled_indexToX (thee.peek(), i);
LPC_Frame_init (lpcframe, predictionOrder);
Sound_into_Sound (sound.peek(), sframe.peek(), t - windowDuration / 2);
Vector_subtractMean (sframe.peek());
Sounds_multiply (sframe.peek(), window.peek());
if (method == LPC_METHOD_AUTO) {
if (! Sound_into_LPC_Frame_auto (sframe.peek(), lpcframe)) {
frameErrorCount++;
}
} else if (method == LPC_METHOD_COVAR) {
if (! Sound_into_LPC_Frame_covar (sframe.peek(), lpcframe)) {
frameErrorCount++;
}
} else if (method == LPC_METHOD_BURG) {
if (! Sound_into_LPC_Frame_burg (sframe.peek(), lpcframe)) {
frameErrorCount++;
}
} else if (method == LPC_METHOD_MARPLE) {
if (! Sound_into_LPC_Frame_marple (sframe.peek(), lpcframe, tol1, tol2)) {
frameErrorCount++;
}
}
if ( (i % 10) == 1) {
Melder_progress ( (double) i / nFrames, L"LPC analysis of frame ",
Melder_integer (i), L" out of ", Melder_integer (nFrames), L".");
}
}
return thee.transfer();
}
static autoSound ComplexSpectrogram_to_Sound2 (ComplexSpectrogram me, double stretchFactor) {
try {
/* original number of samples is odd: imaginary part of last spectral value is zero ->
* phase is either zero or pi
*/
double pi = atan2 (0.0, - 0.5);
double samplingFrequency = 2.0 * my ymax;
double lastFrequency = my y1 + (my ny - 1) * my dy;
int originalNumberOfSamplesProbablyOdd = (my phase [my ny][1] != 0.0 && my phase[my ny][1] != pi) || my ymax - lastFrequency > 0.25 * my dx;
if (my y1 != 0.0) {
Melder_throw (U"A Fourier-transformable Spectrum must have a first frequency of 0 Hz, not ", my y1, U" Hz.");
}
long numberOfSamples = 2 * my ny - (originalNumberOfSamplesProbablyOdd ? 1 : 2 );
double synthesisWindowDuration = numberOfSamples / samplingFrequency;
autoSpectrum spectrum = Spectrum_create (my ymax, my ny);
autoSound synthesisWindow = Sound_createSimple (1, synthesisWindowDuration, samplingFrequency);
long stepSizeSamples = my dx * samplingFrequency * stretchFactor;
double newDuration = (my xmax - my xmin) * stretchFactor + 0.05;
autoSound thee = Sound_createSimple (1, newDuration, samplingFrequency); //TODO
long istart = 1, iend = istart + stepSizeSamples - 1;
for (long iframe = 1; iframe <= my nx; iframe++) {
spectrum -> z[1][1] = sqrt (my z[1][iframe]);
for (long ifreq = 2; ifreq <= my ny; ifreq++) {
double f = my y1 + (ifreq - 1) * my dy;
double a = sqrt (my z[ifreq][iframe]);
double phi = my phase[ifreq][iframe];
double extraPhase = 2.0 * pi * (stretchFactor - 1.0) * my dx * f;
phi += extraPhase;
spectrum -> z[1][ifreq] = a * cos (phi);
spectrum -> z[2][ifreq] = a * sin (phi);
}
autoSound synthesis = Spectrum_to_Sound (spectrum.get());
for (long j = istart; j <= iend; j++) {
thy z[1][j] = synthesis -> z[1][j - istart + 1];
}
istart = iend + 1; iend = istart + stepSizeSamples - 1;
}
return thee;
} catch (MelderError) {
Melder_throw (me, U": no Sound created.");
}
}
autoMelSpectrogram Sound_to_MelSpectrogram (Sound me, double analysisWidth, double dt, double f1_mel, double fmax_mel, double df_mel) {
try {
double t1, samplingFrequency = 1.0 / my dx, nyquist = 0.5 * samplingFrequency;
double windowDuration = 2.0 * analysisWidth; // gaussian window
double fmin_mel = 0.0;
double fbottom = NUMhertzToMel2 (100.0), fceiling = NUMhertzToMel2 (nyquist);
long numberOfFrames;
// Check defaults.
if (fmax_mel <= 0.0 || fmax_mel > fceiling) {
fmax_mel = fceiling;
}
if (fmax_mel <= f1_mel) {
f1_mel = fbottom; fmax_mel = fceiling;
}
if (f1_mel <= 0.0) {
f1_mel = fbottom;
}
if (df_mel <= 0.0) {
df_mel = 100.0;
}
// Determine the number of filters.
long numberOfFilters = lround ((fmax_mel - f1_mel) / df_mel);
fmax_mel = f1_mel + numberOfFilters * df_mel;
Sampled_shortTermAnalysis (me, windowDuration, dt, &numberOfFrames, &t1);
autoSound sframe = Sound_createSimple (1, windowDuration, samplingFrequency);
autoSound window = Sound_createGaussian (windowDuration, samplingFrequency);
autoMelSpectrogram thee = MelSpectrogram_create (my xmin, my xmax, numberOfFrames, dt, t1, fmin_mel, fmax_mel, numberOfFilters, df_mel, f1_mel);
autoMelderProgress progress (U"MelSpectrograms analysis");
for (long iframe = 1; iframe <= numberOfFrames; iframe++) {
double t = Sampled_indexToX (thee.get(), iframe);
Sound_into_Sound (me, sframe.get(), t - windowDuration / 2.0);
Sounds_multiply (sframe.get(), window.get());
Sound_into_MelSpectrogram_frame (sframe.get(), thee.get(), iframe);
if (iframe % 10 == 1) {
Melder_progress ((double) iframe / numberOfFrames, U"Frame ", iframe, U" out of ", numberOfFrames, U".");
}
}
_Spectrogram_windowCorrection ((Spectrogram) thee.get(), window -> nx);
return thee;
} catch (MelderError) {
Melder_throw (me, U": no MelSpectrogram created.");
}
}
autoBarkSpectrogram Sound_to_BarkSpectrogram (Sound me, double analysisWidth, double dt, double f1_bark, double fmax_bark, double df_bark) {
try {
double nyquist = 0.5 / my dx, samplingFrequency = 2 * nyquist;
double windowDuration = 2 * analysisWidth; /* gaussian window */
double zmax = NUMhertzToBark2 (nyquist);
double fmin_bark = 0;
// Check defaults.
if (f1_bark <= 0) {
f1_bark = 1;
}
if (fmax_bark <= 0) {
fmax_bark = zmax;
}
if (df_bark <= 0) {
df_bark = 1;
}
fmax_bark = MIN (fmax_bark, zmax);
long numberOfFilters = lround ( (fmax_bark - f1_bark) / df_bark);
if (numberOfFilters <= 0) {
Melder_throw (U"The combination of filter parameters is not valid.");
}
long numberOfFrames; double t1;
Sampled_shortTermAnalysis (me, windowDuration, dt, & numberOfFrames, & t1);
autoSound sframe = Sound_createSimple (1, windowDuration, samplingFrequency);
autoSound window = Sound_createGaussian (windowDuration, samplingFrequency);
autoBarkSpectrogram thee = BarkSpectrogram_create (my xmin, my xmax, numberOfFrames, dt, t1, fmin_bark, fmax_bark, numberOfFilters, df_bark, f1_bark);
autoMelderProgress progess (U"BarkSpectrogram analysis");
for (long iframe = 1; iframe <= numberOfFrames; iframe++) {
double t = Sampled_indexToX (thee.get(), iframe);
Sound_into_Sound (me, sframe.get(), t - windowDuration / 2.0);
Sounds_multiply (sframe.get(), window.get());
Sound_into_BarkSpectrogram_frame (sframe.get(), thee.get(), iframe);
if (iframe % 10 == 1) {
Melder_progress ( (double) iframe / numberOfFrames, U"BarkSpectrogram analysis: frame ",
iframe, U" from ", numberOfFrames, U".");
}
}
_Spectrogram_windowCorrection ((Spectrogram) thee.get(), window -> nx);
return thee;
} catch (MelderError) {
Melder_throw (me, U": no BarkSpectrogram created.");
}
}
void Artword_Speaker_Sound_movie (Artword artword, Speaker speaker, Sound sound, Graphics graphics) {
try {
static struct playInfo info; // must be static!!!
info. artword = artword;
info. speaker = speaker;
info. graphics = graphics;
autoSound mySound = sound ? nullptr : Sound_createSimple (1, artword -> totalTime, 44100);
Sound_play (sound ? sound : mySound.peek(), playCallback, & info);
} catch (MelderError) {
Melder_throw (artword, U" & ", speaker, U": movie not played.");
}
}
autoCochleagram Sound_to_Cochleagram (Sound me, double dt, double df, double dt_window, double forwardMaskingTime) {
try {
double duration = my nx * my dx;
long nFrames = 1 + (long) floor ((duration - dt_window) / dt);
long nsamp_window = (long) floor (dt_window / my dx), halfnsamp_window = nsamp_window / 2 - 1;
long nf = lround (25.6 / df);
double dampingFactor = forwardMaskingTime > 0.0 ? exp (- dt / forwardMaskingTime) : 0.0; // default 30 ms
double integrationCorrection = 1.0 - dampingFactor;
nsamp_window = halfnsamp_window * 2;
if (nFrames < 2) return autoCochleagram ();
double t1 = my x1 + 0.5 * (duration - my dx - (nFrames - 1) * dt); // centre of first frame
autoCochleagram thee = Cochleagram_create (my xmin, my xmax, nFrames, dt, t1, df, nf);
autoSound window = Sound_createSimple (1, nsamp_window * my dx, 1.0 / my dx);
for (long iframe = 1; iframe <= nFrames; iframe ++) {
double t = Sampled_indexToX (thee.get(), iframe);
long leftSample = Sampled_xToLowIndex (me, t);
long rightSample = leftSample + 1;
long startSample = rightSample - halfnsamp_window;
long endSample = rightSample + halfnsamp_window;
if (startSample < 1) {
Melder_casual (U"Start sample too small: ", startSample,
U" instead of 1.");
startSample = 1;
}
if (endSample > my nx) {
Melder_casual (U"End sample too small: ", endSample,
U" instead of ", my nx,
U".");
endSample = my nx;
}
/* Copy a window to a frame. */
for (long i = 1; i <= nsamp_window; i ++)
window -> z [1] [i] =
( my ny == 1 ? my z[1][i+startSample-1] : 0.5 * (my z[1][i+startSample-1] + my z[2][i+startSample-1]) ) *
(0.5 - 0.5 * cos (2.0 * NUMpi * i / (nsamp_window + 1)));
autoSpectrum spec = Sound_to_Spectrum (window.get(), true);
autoExcitation excitation = Spectrum_to_Excitation (spec.get(), df);
for (long ifreq = 1; ifreq <= nf; ifreq ++)
thy z [ifreq] [iframe] = excitation -> z [1] [ifreq] + ( iframe > 1 ? dampingFactor * thy z [ifreq] [iframe - 1] : 0 );
}
for (long iframe = 1; iframe <= nFrames; iframe ++)
for (long ifreq = 1; ifreq <= nf; ifreq ++)
thy z [ifreq] [iframe] *= integrationCorrection;
return thee;
} catch (MelderError) {
Melder_throw (me, U": not converted to Cochleagram.");
}
}
Cepstrogram Sound_to_Cepstrogram (Sound me, double analysisWidth, double dt, double maximumFrequency) {
try {
double windowDuration = 2 * analysisWidth; /* gaussian window */
long nFrames;
// Convenience: analyse the whole sound into one Cepstrogram_frame
if (windowDuration > my dx * my nx) {
windowDuration = my dx * my nx;
}
double t1, samplingFrequency = 2 * maximumFrequency;
autoSound sound = Sound_resample (me, samplingFrequency, 50);
Sampled_shortTermAnalysis (me, windowDuration, dt, & nFrames, & t1);
autoSound sframe = Sound_createSimple (1, windowDuration, samplingFrequency);
autoSound window = Sound_createGaussian (windowDuration, samplingFrequency);
double qmin, qmax, dq, q1;
long nq;
{ // laziness: find out the proper dimensions
autoSpectrum spec = Sound_to_Spectrum (sframe.peek(), 1);
autoCepstrum cepstrum = Spectrum_to_Cepstrum (spec.peek());
qmin = cepstrum -> xmin; qmax = cepstrum -> xmax; dq = cepstrum -> dx;
q1 = cepstrum -> x1; nq = cepstrum -> nx;
}
autoCepstrogram thee = Cepstrogram_create (my xmin, my xmax, nFrames, dt, t1, qmin, qmax, nq, dq, q1);
autoMelderProgress progress (L"Cepstrogram analysis");
for (long iframe = 1; iframe <= nFrames; iframe++) {
double t = Sampled_indexToX (thee.peek(), iframe);
Sound_into_Sound (sound.peek(), sframe.peek(), t - windowDuration / 2);
Vector_subtractMean (sframe.peek());
Sounds_multiply (sframe.peek(), window.peek());
autoSpectrum spec = Sound_to_Spectrum (sframe.peek(), 1);
autoCepstrum cepstrum = Spectrum_to_Cepstrum (spec.peek());
for (long i = 1; i <= nq; i++) {
thy z[i][iframe] = cepstrum -> z[1][i];
}
if ((iframe % 10) == 1) {
Melder_progress ((double) iframe / nFrames, L"Cepstrogram analysis of frame ",
Melder_integer (iframe), L" out of ", Melder_integer (nFrames), L".");
}
}
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, ": no Cepstrogram created.");
}
}
PowerCepstrogram Sound_to_PowerCepstrogram (Sound me, double pitchFloor, double dt, double maximumFrequency, double preEmphasisFrequency) {
try {
// minimum analysis window has 3 periods of lowest pitch
double analysisWidth = 3 / pitchFloor;
double windowDuration = 2 * analysisWidth; /* gaussian window */
long nFrames;
// Convenience: analyse the whole sound into one Cepstrogram_frame
if (windowDuration > my dx * my nx) {
windowDuration = my dx * my nx;
}
double t1, samplingFrequency = 2 * maximumFrequency;
autoSound sound = Sound_resample (me, samplingFrequency, 50);
Sound_preEmphasis (sound.peek(), preEmphasisFrequency);
Sampled_shortTermAnalysis (me, windowDuration, dt, & nFrames, & t1);
autoSound sframe = Sound_createSimple (1, windowDuration, samplingFrequency);
autoSound window = Sound_createGaussian (windowDuration, samplingFrequency);
// find out the size of the FFT
long nfft = 2;
while (nfft < sframe -> nx) nfft *= 2;
long nq = nfft / 2 + 1;
double qmax = 0.5 * nfft / samplingFrequency, dq = qmax / (nq - 1);
autoPowerCepstrogram thee = PowerCepstrogram_create (my xmin, my xmax, nFrames, dt, t1, 0, qmax, nq, dq, 0);
autoMelderProgress progress (L"Cepstrogram analysis");
for (long iframe = 1; iframe <= nFrames; iframe++) {
double t = Sampled_indexToX (thee.peek(), iframe);
Sound_into_Sound (sound.peek(), sframe.peek(), t - windowDuration / 2);
Vector_subtractMean (sframe.peek());
Sounds_multiply (sframe.peek(), window.peek());
autoSpectrum spec = Sound_to_Spectrum (sframe.peek(), 1); // FFT yes
autoPowerCepstrum cepstrum = Spectrum_to_PowerCepstrum (spec.peek());
for (long i = 1; i <= nq; i++) {
thy z[i][iframe] = cepstrum -> z[1][i];
}
if ((iframe % 10) == 1) {
Melder_progress ((double) iframe / nFrames, L"PowerCepstrogram analysis of frame ",
Melder_integer (iframe), L" out of ", Melder_integer (nFrames), L".");
}
}
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, ": no PowerCepstrogram created.");
}
}
static Sound createGammatone (double midFrequency_Hertz, double samplingFrequency) {
double lengthOfGammatone_seconds = 50.0 / midFrequency_Hertz; // 50 periods
long lengthOfGammatone_samples;
/* EdB's alfa1: */
double latency = 1.95e-3 * pow (midFrequency_Hertz / 1000, -0.725) + 0.6e-3;
/* EdB's beta: */
double decayTime = 1e-3 * pow (midFrequency_Hertz / 1000, -0.663);
/* EdB's omega: */
double midFrequency_radPerSecond = 2 * NUMpi * midFrequency_Hertz;
autoSound gammatone = Sound_createSimple (1, lengthOfGammatone_seconds, samplingFrequency);
lengthOfGammatone_samples = gammatone -> nx;
for (long itime = 1; itime <= lengthOfGammatone_samples; itime ++) {
double time_seconds = (itime - 0.5) / samplingFrequency;
double timeAfterLatency = time_seconds - latency;
double x = timeAfterLatency / decayTime;
if (time_seconds > latency) gammatone -> z [1] [itime] =
x * x * x * exp (- x) * cos (midFrequency_radPerSecond * timeAfterLatency);
}
return gammatone.transfer();
}
static void huber_struct_init (struct huber_struct *hs, double windowDuration,
long p, double samplingFrequency, double location, int wantlocation) {
hs -> w = hs -> work = hs -> a = hs -> c = 0;
hs -> covar = 0; hs -> svd = 0;
hs -> e = Sound_createSimple (1, windowDuration, samplingFrequency);
long n = hs -> e -> nx;
hs -> n = n;
hs -> p = p;
hs -> w = NUMvector<double> (1, n);
hs -> work = NUMvector<double> (1, n);
hs -> a = NUMvector<double> (1, p);
hs -> covar = NUMmatrix<double> (1, p, 1, p);
hs -> c = NUMvector<double> (1, p);
hs -> svd = SVD_create (p, p);
hs -> wantlocation = wantlocation;
if (! wantlocation) {
hs -> location = location;
}
hs -> wantscale = 1;
}
FormantFilter Sound_and_Pitch_to_FormantFilter (Sound me, Pitch thee, double analysisWidth, double dt,
double f1_hz, double fmax_hz, double df_hz, double relative_bw) {
try {
double t1, windowDuration = 2 * analysisWidth; /* gaussian window */
double nyquist = 0.5 / my dx, samplingFrequency = 2 * nyquist, fmin_hz = 0;
long nt, f0_undefined = 0;
if (my xmin > thy xmin || my xmax > thy xmax) Melder_throw
("The domain of the Sound is not included in the domain of the Pitch.");
double f0_median = Pitch_getQuantile (thee, thy xmin, thy xmax, 0.5, kPitch_unit_HERTZ);
if (f0_median == NUMundefined || f0_median == 0) {
f0_median = 100;
Melder_warning (L"Pitch values undefined. Bandwith fixed to 100 Hz. ");
}
if (f1_hz <= 0) {
f1_hz = 100;
}
if (fmax_hz <= 0) {
fmax_hz = nyquist;
}
if (df_hz <= 0) {
df_hz = f0_median / 2;
}
if (relative_bw <= 0) {
relative_bw = 1.1;
}
fmax_hz = MIN (fmax_hz, nyquist);
long nf = floor ( (fmax_hz - f1_hz) / df_hz + 0.5);
Sampled_shortTermAnalysis (me, windowDuration, dt, &nt, &t1);
autoFormantFilter him = FormantFilter_create (my xmin, my xmax, nt, dt, t1,
fmin_hz, fmax_hz, nf, df_hz, f1_hz);
// Temporary objects
autoSound sframe = Sound_createSimple (1, windowDuration, samplingFrequency);
autoSound window = Sound_createGaussian (windowDuration, samplingFrequency);
autoMelderProgress progress (L"Sound & Pitch: To FormantFilter");
for (long i = 1; i <= nt; i++) {
double t = Sampled_indexToX (him.peek(), i);
double b, f0 = Pitch_getValueAtTime (thee, t, kPitch_unit_HERTZ, 0);
if (f0 == NUMundefined || f0 == 0) {
f0_undefined++; f0 = f0_median;
}
b = relative_bw * f0;
Sound_into_Sound (me, sframe.peek(), t - windowDuration / 2);
Sounds_multiply (sframe.peek(), window.peek());
Sound_into_FormantFilter_frame (sframe.peek(), him.peek(), i, b);
if ( (i % 10) == 1) {
Melder_progress ( (double) i / nt, L"Frame ", Melder_integer (i), L" out of ",
Melder_integer (nt), L".");
}
}
double ref = FilterBank_DBREF * gaussian_window_squared_correction (window -> nx);
NUMdmatrix_to_dBs (his z, 1, his ny, 1, his nx, ref, FilterBank_DBFAC, FilterBank_DBFLOOR);
return him.transfer();
} catch (MelderError) {
Melder_throw ("FormantFilter not created from Pitch & FormantFilter.");
}
}
MelFilter Sound_to_MelFilter (Sound me, double analysisWidth, double dt,
double f1_mel, double fmax_mel, double df_mel) {
try {
double t1, samplingFrequency = 1 / my dx, nyquist = 0.5 * samplingFrequency;
double windowDuration = 2 * analysisWidth; /* gaussian window */
double fmin_mel = 0;
double fbottom = HZTOMEL (100.0), fceiling = HZTOMEL (nyquist);
long nt, frameErrorCount = 0;
// Check defaults.
if (fmax_mel <= 0 || fmax_mel > fceiling) {
fmax_mel = fceiling;
}
if (fmax_mel <= f1_mel) {
f1_mel = fbottom; fmax_mel = fceiling;
}
if (f1_mel <= 0) {
f1_mel = fbottom;
}
if (df_mel <= 0) {
df_mel = 100.0;
}
// Determine the number of filters.
long nf = floor ( (fmax_mel - f1_mel) / df_mel + 0.5);
fmax_mel = f1_mel + nf * df_mel;
Sampled_shortTermAnalysis (me, windowDuration, dt, &nt, &t1);
autoSound sframe = Sound_createSimple (1, windowDuration, samplingFrequency);
autoSound window = Sound_createGaussian (windowDuration, samplingFrequency);
autoMelFilter thee = MelFilter_create (my xmin, my xmax, nt, dt, t1, fmin_mel,
fmax_mel, nf, df_mel, f1_mel);
autoMelderProgress progress (L"MelFilters analysis");
for (long i = 1; i <= nt; i++) {
double t = Sampled_indexToX (thee.peek(), i);
Sound_into_Sound (me, sframe.peek(), t - windowDuration / 2);
Sounds_multiply (sframe.peek(), window.peek());
if (! Sound_into_MelFilter_frame (sframe.peek(), thee.peek(), i)) {
frameErrorCount++;
}
if ( (i % 10) == 1) {
Melder_progress ( (double) i / nt, L"Frame ", Melder_integer (i), L" out of ",
Melder_integer (nt), L".");
}
}
if (frameErrorCount) Melder_warning (L"Analysis results of ", Melder_integer (frameErrorCount),
L" frame(s) out of ", Melder_integer (nt), L" will be suspect.");
// Window correction.
double ref = FilterBank_DBREF * gaussian_window_squared_correction (window -> nx);
NUMdmatrix_to_dBs (thy z, 1, thy ny, 1, thy nx, ref, FilterBank_DBFAC, FilterBank_DBFLOOR);
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, ": no MelFilter created.");
}
}
autoSound ComplexSpectrogram_to_Sound (ComplexSpectrogram me, double stretchFactor) {
try {
/* original number of samples is odd: imaginary part of last spectral value is zero ->
* phase is either zero or +/-pi
*/
double pi = atan2 (0.0, - 0.5);
double samplingFrequency = 2.0 * my ymax;
double lastFrequency = my y1 + (my ny - 1) * my dy, lastPhase = my phase[my ny][1];
int originalNumberOfSamplesProbablyOdd = (lastPhase != 0.0 && lastPhase != pi && lastPhase != -pi) ||
my ymax - lastFrequency > 0.25 * my dx;
if (my y1 != 0.0) {
Melder_throw (U"A Fourier-transformable ComplexSpectrogram must have a first frequency of 0 Hz, not ", my y1, U" Hz.");
}
long nsamp_window = 2 * my ny - (originalNumberOfSamplesProbablyOdd ? 1 : 2 );
long halfnsamp_window = nsamp_window / 2;
double synthesisWindowDuration = nsamp_window / samplingFrequency;
autoSpectrum spectrum = Spectrum_create (my ymax, my ny);
autoSound synthesisWindow = Sound_createSimple (1, synthesisWindowDuration, samplingFrequency);
double newDuration = (my xmax - my xmin) * stretchFactor;
autoSound thee = Sound_createSimple (1, newDuration, samplingFrequency); //TODO
double thyStartTime;
for (long iframe = 1; iframe <= my nx; iframe++) {
// "original" sound :
double tmid = Sampled_indexToX (me, iframe);
long leftSample = Sampled_xToLowIndex (thee.get(), tmid);
long rightSample = leftSample + 1;
long startSample = rightSample - halfnsamp_window;
double startTime = Sampled_indexToX (thee.get(), startSample);
if (iframe == 1) {
thyStartTime = Sampled_indexToX (thee.get(), startSample);
}
//long endSample = leftSample + halfnsamp_window;
// New Sound with stretch
long thyStartSample = Sampled_xToLowIndex (thee.get(),thyStartTime);
double thyEndTime = thyStartTime + my dx * stretchFactor;
long thyEndSample = Sampled_xToLowIndex (thee.get(), thyEndTime);
long stretchedStepSizeSamples = thyEndSample - thyStartSample + 1;
//double extraTime = (thyStartSample - startSample + 1) * thy dx;
double extraTime = (thyStartTime - startTime);
spectrum -> z[1][1] = sqrt (my z[1][iframe]);
for (long ifreq = 2; ifreq <= my ny; ifreq++) {
double f = my y1 + (ifreq - 1) * my dy;
double a = sqrt (my z[ifreq][iframe]);
double phi = my phase[ifreq][iframe], intPart;
double extraPhase = 2.0 * pi * modf (extraTime * f, &intPart); // fractional part
phi += extraPhase;
spectrum -> z[1][ifreq] = a * cos (phi);
spectrum -> z[2][ifreq] = a * sin (phi);
}
autoSound synthesis = Spectrum_to_Sound (spectrum.get());
// Where should the sound be placed?
long thyEndSampleP = (long) floor (fmin (thyStartSample + synthesis -> nx - 1, thyStartSample + stretchedStepSizeSamples - 1)); // guard against extreme stretches
if (iframe == my nx) {
thyEndSampleP = (long) floor (fmin (thy nx, thyStartSample + synthesis -> nx - 1)); // ppgb: waarom naar beneden afgerond?
}
for (long j = thyStartSample; j <= thyEndSampleP; j++) {
thy z[1][j] = synthesis -> z[1][j - thyStartSample + 1];
}
thyStartTime += my dx * stretchFactor;
}
return thee;
} catch (MelderError) {
Melder_throw (me, U": no Sound created.");
}
}
Pitch Sound_to_Pitch_shs (Sound me, double timeStep, double minimumPitch,
double maximumFrequency, double ceiling, long maxnSubharmonics, long maxnCandidates,
double compressionFactor, long nPointsPerOctave) {
try {
double firstTime, newSamplingFrequency = 2 * maximumFrequency;
double windowDuration = 2 / minimumPitch, halfWindow = windowDuration / 2;
double atans = nPointsPerOctave * NUMlog2 (65.0 / 50.0) - 1;
// Number of speech samples in the downsampled signal in each frame:
// 100 for windowDuration == 0.04 and newSamplingFrequency == 2500
long nx = lround (windowDuration * newSamplingFrequency);
// The minimum number of points for the fft is 256.
long nfft = 1;
while ( (nfft *= 2) < nx || nfft <= 128) {
;
}
long nfft2 = nfft / 2 + 1;
double frameDuration = nfft / newSamplingFrequency;
double df = newSamplingFrequency / nfft;
// The number of points on the octave scale
double fminl2 = NUMlog2 (minimumPitch), fmaxl2 = NUMlog2 (maximumFrequency);
long nFrequencyPoints = (long) floor ((fmaxl2 - fminl2) * nPointsPerOctave);
double dfl2 = (fmaxl2 - fminl2) / (nFrequencyPoints - 1);
autoSound sound = Sound_resample (me, newSamplingFrequency, 50);
long numberOfFrames;
Sampled_shortTermAnalysis (sound.peek(), windowDuration, timeStep, &numberOfFrames, &firstTime);
autoSound frame = Sound_createSimple (1, frameDuration, newSamplingFrequency);
autoSound hamming = Sound_createHamming (nx / newSamplingFrequency, newSamplingFrequency);
autoPitch thee = Pitch_create (my xmin, my xmax, numberOfFrames, timeStep, firstTime,
ceiling, maxnCandidates);
autoNUMvector<double> cc (1, numberOfFrames);
autoNUMvector<double> specAmp (1, nfft2);
autoNUMvector<double> fl2 (1, nfft2);
autoNUMvector<double> yv2 (1, nfft2);
autoNUMvector<double> arctg (1, nFrequencyPoints);
autoNUMvector<double> al2 (1, nFrequencyPoints);
Melder_assert (frame->nx >= nx);
Melder_assert (hamming->nx == nx);
// Compute the absolute value of the globally largest amplitude w.r.t. the global mean.
double globalMean, globalPeak;
Sound_localMean (sound.peek(), sound -> xmin, sound -> xmax, &globalMean);
Sound_localPeak (sound.peek(), sound -> xmin, sound -> xmax, globalMean, &globalPeak);
/*
For the cubic spline interpolation we need the frequencies on an octave
scale, i.e., a log2 scale. All frequencies must be DIFFERENT, otherwise
the cubic spline interpolation will give corrupt results.
Because log2(f==0) is not defined, we use the heuristic: f[2]-f[1] == f[3]-f[2].
*/
for (long i = 2; i <= nfft2; i++) {
fl2[i] = NUMlog2 ( (i - 1) * df);
}
fl2[1] = 2 * fl2[2] - fl2[3];
// Calculate frequencies regularly spaced on a log2-scale and
// the frequency weighting function.
for (long i = 1; i <= nFrequencyPoints; i++) {
arctg[i] = 0.5 + atan (3 * (i - atans) / nPointsPerOctave) / NUMpi;
}
// Perform the analysis on all frames.
for (long i = 1; i <= numberOfFrames; i++) {
Pitch_Frame pitchFrame = &thy frame[i];
double hm = 1, f0, pitch_strength, localMean, localPeak;
double tmid = Sampled_indexToX (thee.peek(), i); /* The center of this frame */
long nx_tmp = frame -> nx;
// Copy a frame from the sound, apply a hamming window. Get local 'intensity'
frame -> nx = nx; /*begin vies */
Sound_into_Sound (sound.peek(), frame.peek(), tmid - halfWindow);
Sounds_multiply (frame.peek(), hamming.peek());
Sound_localMean (sound.peek(), tmid - 3 * halfWindow, tmid + 3 * halfWindow, &localMean);
Sound_localPeak (sound.peek(), tmid - halfWindow, tmid + halfWindow, localMean, &localPeak);
pitchFrame -> intensity = localPeak > globalPeak ? 1 : localPeak / globalPeak;
frame -> nx = nx_tmp; /* einde vies */
// Get the Fourier spectrum.
autoSpectrum spec = Sound_to_Spectrum (frame.peek(), 1);
Melder_assert (spec->nx == nfft2);
// From complex spectrum to amplitude spectrum.
for (long j = 1; j <= nfft2; j++) {
double rs = spec -> z[1][j], is = spec -> z[2][j];
specAmp[j] = sqrt (rs * rs + is * is);
}
// Enhance the peaks in the spectrum.
spec_enhance_SHS (specAmp.peek(), nfft2);
// Smooth the enhanced spectrum.
spec_smoooth_SHS (specAmp.peek(), nfft2);
// Go to a logarithmic scale and perform cubic spline interpolation to get
// spectral values for the increased number of frequency points.
NUMspline (fl2.peek(), specAmp.peek(), nfft2, 1e30, 1e30, yv2.peek());
for (long j = 1; j <= nFrequencyPoints; j++) {
double f = fminl2 + (j - 1) * dfl2;
NUMsplint (fl2.peek(), specAmp.peek(), yv2.peek(), nfft2, f, &al2[j]);
}
// Multiply by frequency selectivity of the auditory system.
for (long j = 1; j <= nFrequencyPoints; j++) al2[j] = al2[j] > 0 ?
al2[j] * arctg[j] : 0;
// The subharmonic summation. Shift spectra in octaves and sum.
Pitch_Frame_init (pitchFrame, maxnCandidates);
autoNUMvector<double> sumspec (1, nFrequencyPoints);
pitchFrame -> nCandidates = 0; /* !!!!! */
for (long m = 1; m <= maxnSubharmonics + 1; m++) {
long kb = 1 + (long) floor (nPointsPerOctave * NUMlog2 (m));
for (long k = kb; k <= nFrequencyPoints; k++) {
sumspec[k - kb + 1] += al2[k] * hm;
}
hm *= compressionFactor;
}
// First register the voiceless candidate (always present).
Pitch_Frame_addPitch (pitchFrame, 0, 0, maxnCandidates);
/*
Get the best local estimates for the pitch as the maxima of the
subharmonic sum spectrum by parabolic interpolation on three points:
The formula for a parabole with a maximum is:
y(x) = a - b (x - c)^2 with a, b, c >= 0
The three points are (-x, y1), (0, y2) and (x, y3).
The solution for a (the maximum) and c (the position) is:
a = (2 y1 (4 y2 + y3) - y1^2 - (y3 - 4 y2)^2)/( 8 (y1 - 2 y2 + y3)
c = dx (y1 - y3) / (2 (y1 - 2 y2 + y3))
(b = (2 y2 - y1 - y3) / (2 dx^2) )
*/
for (long k = 2; k <= nFrequencyPoints - 1; k++) {
double y1 = sumspec[k - 1], y2 = sumspec[k], y3 = sumspec[k + 1];
if (y2 > y1 && y2 >= y3) {
double denum = y1 - 2 * y2 + y3, tmp = y3 - 4 * y2;
double x = dfl2 * (y1 - y3) / (2 * denum);
double f = pow (2, fminl2 + (k - 1) * dfl2 + x);
double strength = (2 * y1 * (4 * y2 + y3) - y1 * y1 - tmp * tmp) / (8 * denum);
Pitch_Frame_addPitch (pitchFrame, f, strength, maxnCandidates);
}
}
/*
Check whether f0 corresponds to an actual periodicity T = 1 / f0:
correlate two signal periods of duration T, one starting at the
middle of the interval and one starting T seconds before.
If there is periodicity the correlation coefficient should be high.
However, some sounds do not show any regularity, or very low
frequency and regularity, and nevertheless have a definite
pitch, e.g. Shepard sounds.
*/
Pitch_Frame_getPitch (pitchFrame, &f0, &pitch_strength);
if (f0 > 0) {
cc[i] = Sound_correlateParts (sound.peek(), tmid - 1.0 / f0, tmid, 1.0 / f0);
}
}
// Base V/UV decision on correlation coefficients.
// Resize the pitch strengths w.r.t. the cc.
double vuvCriterium = 0.52;
for (long i = 1; i <= numberOfFrames; i++) {
Pitch_Frame_resizeStrengths (& thy frame[i], cc[i], vuvCriterium);
}
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, U": no Pitch (shs) created.");
}
}
Sound Artword_Speaker_to_Sound (Artword artword, Speaker speaker,
double fsamp, int oversampling,
Sound *out_w1, int iw1, Sound *out_w2, int iw2, Sound *out_w3, int iw3,
Sound *out_p1, int ip1, Sound *out_p2, int ip2, Sound *out_p3, int ip3,
Sound *out_v1, int iv1, Sound *out_v2, int iv2, Sound *out_v3, int iv3)
{
try {
autoSound result = Sound_createSimple (1, artword -> totalTime, fsamp);
long numberOfSamples = result -> nx;
double minTract [1+78], maxTract [1+78]; /* For drawing. */
double Dt = 1 / fsamp / oversampling,
rho0 = 1.14,
c = 353,
onebyc2 = 1.0 / (c * c),
rho0c2 = rho0 * c * c,
halfDt = 0.5 * Dt,
twoDt = 2 * Dt,
halfc2Dt = 0.5 * c * c * Dt,
twoc2Dt = 2 * c * c * Dt,
onebytworho0 = 1.0 / (2.0 * rho0),
Dtbytworho0 = Dt / (2.0 * rho0);
double tension, rrad, onebygrad, totalVolume;
autoArt art = Art_create ();
long sample;
int n, m, M;
autoDelta delta = Speaker_to_Delta (speaker);
autoMelderMonitor monitor (U"Articulatory synthesis");
Artword_intoArt (artword, art.peek(), 0.0);
Art_Speaker_intoDelta (art.peek(), speaker, delta.peek());
M = delta -> numberOfTubes;
autoSound w1, w2, w3, p1, p2, p3, v1, v2, v3;
if (iw1 > 0 && iw1 <= M) w1.reset (Sound_createSimple (1, artword -> totalTime, fsamp)); else iw1 = 0;
if (iw2 > 0 && iw2 <= M) w2.reset (Sound_createSimple (1, artword -> totalTime, fsamp)); else iw2 = 0;
if (iw3 > 0 && iw3 <= M) w3.reset (Sound_createSimple (1, artword -> totalTime, fsamp)); else iw3 = 0;
if (ip1 > 0 && ip1 <= M) p1.reset (Sound_createSimple (1, artword -> totalTime, fsamp)); else ip1 = 0;
if (ip2 > 0 && ip2 <= M) p2.reset (Sound_createSimple (1, artword -> totalTime, fsamp)); else ip2 = 0;
if (ip3 > 0 && ip3 <= M) p3.reset (Sound_createSimple (1, artword -> totalTime, fsamp)); else ip3 = 0;
if (iv1 > 0 && iv1 <= M) v1.reset (Sound_createSimple (1, artword -> totalTime, fsamp)); else iv1 = 0;
if (iv2 > 0 && iv2 <= M) v2.reset (Sound_createSimple (1, artword -> totalTime, fsamp)); else iv2 = 0;
if (iv3 > 0 && iv3 <= M) v3.reset (Sound_createSimple (1, artword -> totalTime, fsamp)); else iv3 = 0;
/* Initialize drawing. */
{ int i; for (i = 1; i <= 78; i ++) { minTract [i] = 100; maxTract [i] = -100; } }
totalVolume = 0.0;
for (m = 1; m <= M; m ++) {
Delta_Tube t = delta->tube + m;
if (! t -> left1 && ! t -> right1) continue;
t->Dx = t->Dxeq; t->dDxdt = 0; /* 5.113 */
t->Dy = t->Dyeq; t->dDydt = 0; /* 5.113 */
t->Dz = t->Dzeq; /* 5.113 */
t->A = t->Dz * ( t->Dy >= t->dy ? t->Dy + Dymin :
t->Dy <= - t->dy ? Dymin :
(t->dy + t->Dy) * (t->dy + t->Dy) / (4 * t->dy) + Dymin ); /* 4.4, 4.5 */
#if EQUAL_TUBE_WIDTHS
t->A = 0.0001;
#endif
t->Jleft = t->Jright = 0; /* 5.113 */
t->Qleft = t->Qright = rho0c2; /* 5.113 */
t->pleft = t->pright = 0; /* 5.114 */
t->Kleft = t->Kright = 0; /* 5.114 */
t->V = t->A * t->Dx; /* 5.114 */
totalVolume += t->V;
}
//Melder_casual (U"Starting volume: ", totalVolume * 1000, U" litres.");
for (sample = 1; sample <= numberOfSamples; sample ++) {
double time = (sample - 1) / fsamp;
Artword_intoArt (artword, art.peek(), time);
Art_Speaker_intoDelta (art.peek(), speaker, delta.peek());
if (sample % MONITOR_SAMPLES == 0 && monitor.graphics()) { // because we can be in batch
Graphics graphics = monitor.graphics();
double area [1+78];
Graphics_Viewport vp;
for (int i = 1; i <= 78; i ++) {
area [i] = delta -> tube [i]. A;
if (area [i] < minTract [i]) minTract [i] = area [i];
if (area [i] > maxTract [i]) maxTract [i] = area [i];
}
Graphics_clearWs (graphics);
vp = Graphics_insetViewport (monitor.graphics(), 0, 0.5, 0.5, 1);
Graphics_setWindow (graphics, 0, 1, 0, 0.05);
Graphics_setColour (graphics, Graphics_RED);
Graphics_function (graphics, minTract, 1, 35, 0, 0.9);
Graphics_function (graphics, maxTract, 1, 35, 0, 0.9);
Graphics_setColour (graphics, Graphics_BLACK);
Graphics_function (graphics, area, 1, 35, 0, 0.9);
Graphics_setLineType (graphics, Graphics_DOTTED);
Graphics_line (graphics, 0, 0, 1, 0);
Graphics_setLineType (graphics, Graphics_DRAWN);
Graphics_resetViewport (graphics, vp);
vp = Graphics_insetViewport (graphics, 0, 0.5, 0, 0.5);
Graphics_setWindow (graphics, 0, 1, -0.000003, 0.00001);
Graphics_setColour (graphics, Graphics_RED);
Graphics_function (graphics, minTract, 36, 37, 0.2, 0.8);
Graphics_function (graphics, maxTract, 36, 37, 0.2, 0.8);
Graphics_setColour (graphics, Graphics_BLACK);
Graphics_function (graphics, area, 36, 37, 0.2, 0.8);
Graphics_setLineType (graphics, Graphics_DOTTED);
Graphics_line (graphics, 0, 0, 1, 0);
Graphics_setLineType (graphics, Graphics_DRAWN);
Graphics_resetViewport (graphics, vp);
vp = Graphics_insetViewport (graphics, 0.5, 1, 0.5, 1);
Graphics_setWindow (graphics, 0, 1, 0, 0.001);
Graphics_setColour (graphics, Graphics_RED);
Graphics_function (graphics, minTract, 38, 64, 0, 1);
Graphics_function (graphics, maxTract, 38, 64, 0, 1);
Graphics_setColour (graphics, Graphics_BLACK);
Graphics_function (graphics, area, 38, 64, 0, 1);
Graphics_setLineType (graphics, Graphics_DOTTED);
Graphics_line (graphics, 0, 0, 1, 0);
Graphics_setLineType (graphics, Graphics_DRAWN);
Graphics_resetViewport (graphics, vp);
vp = Graphics_insetViewport (graphics, 0.5, 1, 0, 0.5);
Graphics_setWindow (graphics, 0, 1, 0.001, 0);
Graphics_setColour (graphics, Graphics_RED);
Graphics_function (graphics, minTract, 65, 78, 0.5, 1);
Graphics_function (graphics, maxTract, 65, 78, 0.5, 1);
Graphics_setColour (graphics, Graphics_BLACK);
Graphics_function (graphics, area, 65, 78, 0.5, 1);
Graphics_setLineType (graphics, Graphics_DRAWN);
Graphics_resetViewport (graphics, vp);
Melder_monitor ((double) sample / numberOfSamples, U"Articulatory synthesis: ", Melder_half (time), U" seconds");
}
for (n = 1; n <= oversampling; n ++) {
for (m = 1; m <= M; m ++) {
Delta_Tube t = delta -> tube + m;
if (! t -> left1 && ! t -> right1) continue;
/* New geometry. */
#if CONSTANT_TUBE_LENGTHS
t->Dxnew = t->Dx;
#else
t->dDxdtnew = (t->dDxdt + Dt * 10000 * (t->Dxeq - t->Dx)) /
(1 + 200 * Dt); /* Critical damping, 10 ms. */
t->Dxnew = t->Dx + t->dDxdtnew * Dt;
#endif
/* 3-way: equal lengths. */
/* This requires left tubes to be processed before right tubes. */
if (t->left1 && t->left1->right2) t->Dxnew = t->left1->Dxnew;
t->Dz = t->Dzeq; /* immediate... */
t->eleft = (t->Qleft - t->Kleft) * t->V; /* 5.115 */
t->eright = (t->Qright - t->Kright) * t->V; /* 5.115 */
t->e = 0.5 * (t->eleft + t->eright); /* 5.116 */
t->p = 0.5 * (t->pleft + t->pright); /* 5.116 */
t->DeltaP = t->e / t->V - rho0c2; /* 5.117 */
t->v = t->p / (rho0 + onebyc2 * t->DeltaP); /* 5.118 */
{
double dDy = t->Dyeq - t->Dy;
double cubic = t->k3 * dDy * dDy;
Delta_Tube l1 = t->left1, l2 = t->left2, r1 = t->right1, r2 = t->right2;
tension = dDy * (t->k1 + cubic);
t->B = 2 * t->Brel * sqrt (t->mass * (t->k1 + 3 * cubic));
if (t->k1left1 != 0.0 && l1)
tension += t->k1left1 * t->k1 * (dDy - (l1->Dyeq - l1->Dy));
if (t->k1left2 != 0.0 && l2)
tension += t->k1left2 * t->k1 * (dDy - (l2->Dyeq - l2->Dy));
if (t->k1right1 != 0.0 && r1)
tension += t->k1right1 * t->k1 * (dDy - (r1->Dyeq - r1->Dy));
if (t->k1right2 != 0.0 && r2)
tension += t->k1right2 * t->k1 * (dDy - (r2->Dyeq - r2->Dy));
}
if (t->Dy < t->dy) {
if (t->Dy >= - t->dy) {
double dDy = t->dy - t->Dy, dDy2 = dDy * dDy;
tension += dDy2 / (4 * t->dy) * (t->s1 + 0.5 * t->s3 * dDy2);
t->B += 2 * dDy / (2 * t->dy) *
sqrt (t->mass * (t->s1 + t->s3 * dDy2));
} else {
tension -= t->Dy * (t->s1 + t->s3 * (t->Dy * t->Dy + t->dy * t->dy));
t->B += 2 * sqrt (t->mass * (t->s1 + t->s3 * (3 * t->Dy * t->Dy + t->dy * t->dy)));
}
}
t->dDydtnew = (t->dDydt + Dt / t->mass * (tension + 2 * t->DeltaP * t->Dz * t->Dx)) /
(1 + t->B * Dt / t->mass); /* 5.119 */
t->Dynew = t->Dy + t->dDydtnew * Dt; /* 5.119 */
#if NO_MOVING_WALLS
t->Dynew = t->Dy;
#endif
t->Anew = t->Dz * ( t->Dynew >= t->dy ? t->Dynew + Dymin :
t->Dynew <= - t->dy ? Dymin :
(t->dy + t->Dynew) * (t->dy + t->Dynew) / (4 * t->dy) + Dymin ); /* 4.4, 4.5 */
#if EQUAL_TUBE_WIDTHS
t->Anew = 0.0001;
#endif
t->Ahalf = 0.5 * (t->A + t->Anew); /* 5.120 */
t->Dxhalf = 0.5 * (t->Dxnew + t->Dx); /* 5.121 */
t->Vnew = t->Anew * t->Dxnew; /* 5.128 */
{ double oneByDyav = t->Dz / t->A;
/*t->R = 12 * 1.86e-5 * t->parallel * t->parallel * oneByDyav * oneByDyav;*/
if (t->Dy < 0)
t->R = 12 * 1.86e-5 / (Dymin * Dymin + t->dy * t->dy);
else
t->R = 12 * 1.86e-5 * t->parallel * t->parallel /
((t->Dy + Dymin) * (t->Dy + Dymin) + t->dy * t->dy);
t->R += 0.3 * t->parallel * oneByDyav; /* 5.23 */ }
t->r = (1 + t->R * Dt / rho0) * t->Dxhalf / t->Anew; /* 5.122 */
t->ehalf = t->e + halfc2Dt * (t->Jleft - t->Jright); /* 5.123 */
t->phalf = (t->p + halfDt * (t->Qleft - t->Qright) / t->Dx) / (1 + Dtbytworho0 * t->R); /* 5.123 */
#if MASS_LEAPFROG
t->ehalf = t->ehalfold + 2 * halfc2Dt * (t->Jleft - t->Jright);
#endif
t->Jhalf = t->phalf * t->Ahalf; /* 5.124 */
t->Qhalf = t->ehalf / (t->Ahalf * t->Dxhalf) + onebytworho0 * t->phalf * t->phalf; /* 5.124 */
#if NO_BERNOULLI_EFFECT
t->Qhalf = t->ehalf / (t->Ahalf * t->Dxhalf);
#endif
}
for (m = 1; m <= M; m ++) { /* Compute Jleftnew and Qleftnew. */
Delta_Tube l = delta->tube + m, r1 = l -> right1, r2 = l -> right2, r = r1;
Delta_Tube l1 = l, l2 = r ? r -> left2 : NULL;
if (l->left1 == NULL) { /* Closed boundary at the left side (diaphragm)? */
if (r == NULL) continue; /* Tube not connected at all. */
l->Jleftnew = 0; /* 5.132. */
l->Qleftnew = (l->eleft - twoc2Dt * l->Jhalf) / l->Vnew; /* 5.132. */
}
else /* Left boundary open to another tube will be handled... */
(void) 0; /* ...together with the right boundary of the tube to the left. */
if (r == NULL) { /* Open boundary at the right side (lips, nostrils)? */
rrad = 1 - c * Dt / 0.02; /* Radiation resistance, 5.135. */
onebygrad = 1 / (1 + c * Dt / 0.02); /* Radiation conductance, 5.135. */
#if NO_RADIATION_DAMPING
rrad = 0;
onebygrad = 0;
#endif
l->prightnew = ((l->Dxhalf / Dt + c * onebygrad) * l->pright +
2 * ((l->Qhalf - rho0c2) - (l->Qright - rho0c2) * onebygrad)) /
(l->r * l->Anew / Dt + c * onebygrad); /* 5.136 */
l->Jrightnew = l->prightnew * l->Anew; /* 5.136 */
l->Qrightnew = (rrad * (l->Qright - rho0c2) +
c * (l->prightnew - l->pright)) * onebygrad + rho0c2; /* 5.136 */
} else if (l2 == NULL && r2 == NULL) { /* Two-way boundary. */
if (l->v > criticalVelocity && l->A < r->A) {
l->Pturbrightnew = -0.5 * rho0 * (l->v - criticalVelocity) *
(1 - l->A / r->A) * (1 - l->A / r->A) * l->v;
if (l->Pturbrightnew != 0.0)
l->Pturbrightnew *= 1 + NUMrandomGauss (0, noiseFactor) /* * l->A */;
}
if (r->v < - criticalVelocity && r->A < l->A) {
l->Pturbrightnew = 0.5 * rho0 * (r->v + criticalVelocity) *
(1 - r->A / l->A) * (1 - r->A / l->A) * r->v;
if (l->Pturbrightnew != 0.0)
l->Pturbrightnew *= 1 + NUMrandomGauss (0, noiseFactor) /* * r->A */;
}
#if NO_TURBULENCE
l->Pturbrightnew = 0;
#endif
l->Jrightnew = r->Jleftnew =
(l->Dxhalf * l->pright + r->Dxhalf * r->pleft +
twoDt * (l->Qhalf - r->Qhalf + l->Pturbright)) /
(l->r + r->r); /* 5.127 */
#if B91
l->Jrightnew = r->Jleftnew =
(l->pright + r->pleft +
2 * twoDt * (l->Qhalf - r->Qhalf + l->Pturbright) / (l->Dxhalf + r->Dxhalf)) /
(l->r / l->Dxhalf + r->r / r->Dxhalf);
#endif
l->prightnew = l->Jrightnew / l->Anew; /* 5.128 */
r->pleftnew = r->Jleftnew / r->Anew; /* 5.128 */
l->Krightnew = onebytworho0 * l->prightnew * l->prightnew; /* 5.128 */
r->Kleftnew = onebytworho0 * r->pleftnew * r->pleftnew; /* 5.128 */
#if NO_BERNOULLI_EFFECT
l->Krightnew = r->Kleftnew = 0;
#endif
l->Qrightnew =
(l->eright + r->eleft + twoc2Dt * (l->Jhalf - r->Jhalf)
+ l->Krightnew * l->Vnew + (r->Kleftnew - l->Pturbrightnew) * r->Vnew) /
(l->Vnew + r->Vnew); /* 5.131 */
r->Qleftnew = l->Qrightnew + l->Pturbrightnew; /* 5.131 */
} else if (r2) { /* Two adjacent tubes at the right side (velic). */
r1->Jleftnew =
(r1->Jleft * r1->Dxhalf * (1 / (l->A + r2->A) + 1 / r1->A) +
twoDt * ((l->Ahalf * l->Qhalf + r2->Ahalf * r2->Qhalf ) / (l->Ahalf + r2->Ahalf) - r1->Qhalf)) /
(1 / (1 / l->r + 1 / r2->r) + r1->r); /* 5.138 */
r2->Jleftnew =
(r2->Jleft * r2->Dxhalf * (1 / (l->A + r1->A) + 1 / r2->A) +
twoDt * ((l->Ahalf * l->Qhalf + r1->Ahalf * r1->Qhalf ) / (l->Ahalf + r1->Ahalf) - r2->Qhalf)) /
(1 / (1 / l->r + 1 / r1->r) + r2->r); /* 5.138 */
l->Jrightnew = r1->Jleftnew + r2->Jleftnew; /* 5.139 */
l->prightnew = l->Jrightnew / l->Anew; /* 5.128 */
r1->pleftnew = r1->Jleftnew / r1->Anew; /* 5.128 */
r2->pleftnew = r2->Jleftnew / r2->Anew; /* 5.128 */
l->Krightnew = onebytworho0 * l->prightnew * l->prightnew; /* 5.128 */
r1->Kleftnew = onebytworho0 * r1->pleftnew * r1->pleftnew; /* 5.128 */
r2->Kleftnew = onebytworho0 * r2->pleftnew * r2->pleftnew; /* 5.128 */
#if NO_BERNOULLI_EFFECT
l->Krightnew = r1->Kleftnew = r2->Kleftnew = 0;
#endif
l->Qrightnew = r1->Qleftnew = r2->Qleftnew =
(l->eright + r1->eleft + r2->eleft + twoc2Dt * (l->Jhalf - r1->Jhalf - r2->Jhalf) +
l->Krightnew * l->Vnew + r1->Kleftnew * r1->Vnew + r2->Kleftnew * r2->Vnew) /
(l->Vnew + r1->Vnew + r2->Vnew); /* 5.137 */
} else {
Melder_assert (l2 != NULL);
l1->Jrightnew =
(l1->Jright * l1->Dxhalf * (1 / (r->A + l2->A) + 1 / l1->A) -
twoDt * ((r->Ahalf * r->Qhalf + l2->Ahalf * l2->Qhalf ) / (r->Ahalf + l2->Ahalf) - l1->Qhalf)) /
(1 / (1 / r->r + 1 / l2->r) + l1->r); /* 5.138 */
l2->Jrightnew =
(l2->Jright * l2->Dxhalf * (1 / (r->A + l1->A) + 1 / l2->A) -
twoDt * ((r->Ahalf * r->Qhalf + l1->Ahalf * l1->Qhalf ) / (r->Ahalf + l1->Ahalf) - l2->Qhalf)) /
(1 / (1 / r->r + 1 / l1->r) + l2->r); /* 5.138 */
r->Jleftnew = l1->Jrightnew + l2->Jrightnew; /* 5.139 */
r->pleftnew = r->Jleftnew / r->Anew; /* 5.128 */
l1->prightnew = l1->Jrightnew / l1->Anew; /* 5.128 */
l2->prightnew = l2->Jrightnew / l2->Anew; /* 5.128 */
r->Kleftnew = onebytworho0 * r->pleftnew * r->pleftnew; /* 5.128 */
l1->Krightnew = onebytworho0 * l1->prightnew * l1->prightnew; /* 5.128 */
l2->Krightnew = onebytworho0 * l2->prightnew * l2->prightnew; /* 5.128 */
#if NO_BERNOULLI_EFFECT
r->Kleftnew = l1->Krightnew = l2->Krightnew = 0;
#endif
r->Qleftnew = l1->Qrightnew = l2->Qrightnew =
(r->eleft + l1->eright + l2->eright + twoc2Dt * (l1->Jhalf + l2->Jhalf - r->Jhalf) +
r->Kleftnew * r->Vnew + l1->Krightnew * l1->Vnew + l2->Krightnew * l2->Vnew) /
(r->Vnew + l1->Vnew + l2->Vnew); /* 5.137 */
}
}
/* Save some results. */
if (n == (oversampling + 1) / 2) {
double out = 0.0;
for (m = 1; m <= M; m ++) {
Delta_Tube t = delta->tube + m;
out += rho0 * t->Dx * t->Dz * t->dDydt * Dt * 1000; /* Radiation of wall movement, 5.140. */
if (t->right1 == NULL)
out += t->Jrightnew - t->Jright; /* Radiation of open tube end. */
}
result -> z [1] [sample] = out /= 4 * NUMpi * 0.4 * Dt; /* At 0.4 metres. */
if (iw1) w1 -> z [1] [sample] = delta->tube[iw1].Dy;
if (iw2) w2 -> z [1] [sample] = delta->tube[iw2].Dy;
if (iw3) w3 -> z [1] [sample] = delta->tube[iw3].Dy;
if (ip1) p1 -> z [1] [sample] = delta->tube[ip1].DeltaP;
if (ip2) p2 -> z [1] [sample] = delta->tube[ip2].DeltaP;
if (ip3) p3 -> z [1] [sample] = delta->tube[ip3].DeltaP;
if (iv1) v1 -> z [1] [sample] = delta->tube[iv1].v;
if (iv2) v2 -> z [1] [sample] = delta->tube[iv2].v;
if (iv3) v3 -> z [1] [sample] = delta->tube[iv3].v;
}
for (m = 1; m <= M; m ++) {
Delta_Tube t = delta->tube + m;
t->Jleft = t->Jleftnew;
t->Jright = t->Jrightnew;
t->Qleft = t->Qleftnew;
t->Qright = t->Qrightnew;
t->Dy = t->Dynew;
t->dDydt = t->dDydtnew;
t->A = t->Anew;
t->Dx = t->Dxnew;
t->dDxdt = t->dDxdtnew;
t->eleft = t->eleftnew;
t->eright = t->erightnew;
#if MASS_LEAPFROG
t->ehalfold = t->ehalf;
#endif
t->pleft = t->pleftnew;
t->pright = t->prightnew;
t->Kleft = t->Kleftnew;
t->Kright = t->Krightnew;
t->V = t->Vnew;
t->Pturbright = t->Pturbrightnew;
}
}
}
totalVolume = 0.0;
for (m = 1; m <= M; m ++)
totalVolume += delta->tube [m]. V;
//Melder_casual (U"Ending volume: ", totalVolume * 1000, U" litres.");
if (out_w1) *out_w1 = w1.transfer();
if (out_w2) *out_w2 = w2.transfer();
if (out_w3) *out_w3 = w3.transfer();
if (out_p1) *out_p1 = p1.transfer();
if (out_p2) *out_p2 = p2.transfer();
if (out_p3) *out_p3 = p3.transfer();
if (out_v1) *out_v1 = v1.transfer();
if (out_v2) *out_v2 = v2.transfer();
if (out_v3) *out_v3 = v3.transfer();
return result.transfer();
} catch (MelderError) {
Melder_throw (artword, U" & ", speaker, U": articulatory synthesis not performed.");
}
}
SPINET Sound_to_SPINET (Sound me, double timeStep, double windowDuration,
double minimumFrequencyHz, double maximumFrequencyHz, long nFilters,
double excitationErbProportion, double inhibitionErbProportion)
{
Sound window = NULL, frame = NULL; SPINET thee = NULL;
long i, j, k, numberOfFrames;
double firstTime, b = 1.02, samplingFrequency = 1 / my dx;
double *f = NULL, *bw = NULL, *aex = NULL, *ain = NULL;
if (timeStep < my dx) timeStep = my dx;
if (maximumFrequencyHz > samplingFrequency / 2) maximumFrequencyHz = samplingFrequency / 2;
if (! Sampled_shortTermAnalysis (me, windowDuration, timeStep, &numberOfFrames, &firstTime) ||
! (thee = SPINET_create (my xmin, my xmax, numberOfFrames, timeStep, firstTime,
minimumFrequencyHz, maximumFrequencyHz, nFilters,
excitationErbProportion, inhibitionErbProportion)) ||
! (window = Sound_createGaussian (windowDuration, samplingFrequency)) ||
! (frame = Sound_createSimple (1, windowDuration, samplingFrequency)) ||
! (f = NUMdvector (1, nFilters)) || ! (bw = NUMdvector (1, nFilters)) ||
! (aex = NUMdvector (1, nFilters)) || ! (ain = NUMdvector (1, nFilters))) goto cleanup;
/*
Cochlear filterbank: gammatone
*/
for (i=1; i <= nFilters; i++)
{
f[i] = NUMerbToHertz (thy y1 + (i - 1) * thy dy);
bw[i] = 2 * NUMpi * b * (f[i] * (6.23e-6 * f[i] + 93.39e-3) + 28.52);
}
Melder_progress1 (0.0, L"SPINET analysis");
for (i=1; i <= nFilters; i++)
{
Sound gammaTone = NULL, filtered = NULL;
/* Contribution of outer & middle ear and phase locking */
double bb = (f[i] / 1000) * exp (- f[i] / 1000);
/* Time where gammafunction envelope has its maximum */
double tgammaMax = (thy gamma - 1) / bw[i];
/* Amplitude at tgammaMax */
double gammaMaxAmplitude = pow ((thy gamma - 1) / (NUMe * bw[i]), (thy gamma - 1));
double timeCorrection = tgammaMax - windowDuration / 2;
if (! (gammaTone = Sound_createGammaTone (0, 0.1, samplingFrequency,
thy gamma, b, f[i], 0, 0, 0)) ||
/* filtering can be made 30% faster by taking Spectrum(me) outside the loop */
! (filtered = Sounds_convolve (me, gammaTone, kSounds_convolve_scaling_SUM, kSounds_convolve_signalOutsideTimeDomain_ZERO))) { forget (gammaTone); goto cleanup; }
/*
To energy measure: weigh with broad-band transfer function
*/
for (j=1; j <= numberOfFrames; j++)
{
Sound_into_Sound (filtered, frame, Sampled_indexToX (thee, j) + timeCorrection);
Sounds_multiply (frame, window);
thy y[i][j] = Sound_power (frame) * bb / gammaMaxAmplitude;
}
forget (filtered); forget (gammaTone);
if (! Melder_progress5 ((double)i / nFilters, L"SPINET: filter ", Melder_integer (i), L" from ",
Melder_integer (nFilters), L".")) goto cleanup;
}
/*
Excitatory and inhibitory area functions
*/
for (i=1; i <= nFilters; i++)
{
for (k=1; k <= nFilters; k++)
{
double fr = (f[k] - f[i]) / bw[i];
aex[i] += fgamma (fr / thy excitationErbProportion, thy gamma);
ain[i] += fgamma (fr / thy inhibitionErbProportion, thy gamma);
}
}
/*
On-center off-surround interactions
*/
for (j=1; j <= numberOfFrames; j++)
for (i=1; i <= nFilters; i++)
{
double a = 0;
for (k=1; k <= nFilters; k++)
{
double fr = (f[k] - f[i]) / bw[i];
double hexsq = fgamma (fr / thy excitationErbProportion, thy gamma);
double hinsq = fgamma (fr / thy inhibitionErbProportion, thy gamma);
a += thy y[k][j] * (hexsq / aex[i] - hinsq / ain[i]);
}
thy s[i][j] = a > 0 ? a : 0;
}
Melder_progress1 (1.0, NULL);
cleanup:
NUMdvector_free (aex, 1); NUMdvector_free (ain, 1);
NUMdvector_free (f, 1); NUMdvector_free (bw, 1);
forget (window); forget (frame);
if (! Melder_hasError()) return thee;
forget (thee);
return Melder_errorp1 (L"Sound_to_SPINET: not performed.");
}
autoEEG EEG_readFromBdfFile (MelderFile file) {
try {
autofile f = Melder_fopen (file, "rb");
char buffer [81];
fread (buffer, 1, 8, f); buffer [8] = '\0';
bool is24bit = buffer [0] == (char) 255;
fread (buffer, 1, 80, f); buffer [80] = '\0';
trace (U"Local subject identification: \"", Melder_peek8to32 (buffer), U"\"");
fread (buffer, 1, 80, f); buffer [80] = '\0';
trace (U"Local recording identification: \"", Melder_peek8to32 (buffer), U"\"");
fread (buffer, 1, 8, f); buffer [8] = '\0';
trace (U"Start date of recording: \"", Melder_peek8to32 (buffer), U"\"");
fread (buffer, 1, 8, f); buffer [8] = '\0';
trace (U"Start time of recording: \"", Melder_peek8to32 (buffer), U"\"");
fread (buffer, 1, 8, f); buffer [8] = '\0';
long numberOfBytesInHeaderRecord = atol (buffer);
trace (U"Number of bytes in header record: ", numberOfBytesInHeaderRecord);
fread (buffer, 1, 44, f); buffer [44] = '\0';
trace (U"Version of data format: \"", Melder_peek8to32 (buffer), U"\"");
fread (buffer, 1, 8, f); buffer [8] = '\0';
long numberOfDataRecords = strtol (buffer, nullptr, 10);
trace (U"Number of data records: ", numberOfDataRecords);
fread (buffer, 1, 8, f); buffer [8] = '\0';
double durationOfDataRecord = atof (buffer);
trace (U"Duration of a data record: ", durationOfDataRecord);
fread (buffer, 1, 4, f); buffer [4] = '\0';
long numberOfChannels = atol (buffer);
trace (U"Number of channels in data record: ", numberOfChannels);
if (numberOfBytesInHeaderRecord != (numberOfChannels + 1) * 256)
Melder_throw (U"Number of bytes in header record (", numberOfBytesInHeaderRecord,
U") doesn't match number of channels (", numberOfChannels, U").");
autostring32vector channelNames (1, numberOfChannels);
for (long ichannel = 1; ichannel <= numberOfChannels; ichannel ++) {
fread (buffer, 1, 16, f); buffer [16] = '\0'; // labels of the channels
/*
* Strip all final spaces.
*/
for (int i = 15; i >= 0; i --) {
if (buffer [i] == ' ') {
buffer [i] = '\0';
} else {
break;
}
}
channelNames [ichannel] = Melder_8to32 (buffer);
trace (U"Channel <<", channelNames [ichannel], U">>");
}
bool hasLetters = str32equ (channelNames [numberOfChannels], U"EDF Annotations");
double samplingFrequency = NUMundefined;
for (long channel = 1; channel <= numberOfChannels; channel ++) {
fread (buffer, 1, 80, f); buffer [80] = '\0'; // transducer type
}
for (long channel = 1; channel <= numberOfChannels; channel ++) {
fread (buffer, 1, 8, f); buffer [8] = '\0'; // physical dimension of channels
}
autoNUMvector <double> physicalMinimum (1, numberOfChannels);
for (long ichannel = 1; ichannel <= numberOfChannels; ichannel ++) {
fread (buffer, 1, 8, f); buffer [8] = '\0';
physicalMinimum [ichannel] = atof (buffer);
}
autoNUMvector <double> physicalMaximum (1, numberOfChannels);
for (long ichannel = 1; ichannel <= numberOfChannels; ichannel ++) {
fread (buffer, 1, 8, f); buffer [8] = '\0';
physicalMaximum [ichannel] = atof (buffer);
}
autoNUMvector <double> digitalMinimum (1, numberOfChannels);
for (long ichannel = 1; ichannel <= numberOfChannels; ichannel ++) {
fread (buffer, 1, 8, f); buffer [8] = '\0';
digitalMinimum [ichannel] = atof (buffer);
}
autoNUMvector <double> digitalMaximum (1, numberOfChannels);
for (long ichannel = 1; ichannel <= numberOfChannels; ichannel ++) {
fread (buffer, 1, 8, f); buffer [8] = '\0';
digitalMaximum [ichannel] = atof (buffer);
}
for (long channel = 1; channel <= numberOfChannels; channel ++) {
fread (buffer, 1, 80, f); buffer [80] = '\0'; // prefiltering
}
long numberOfSamplesPerDataRecord = 0;
for (long channel = 1; channel <= numberOfChannels; channel ++) {
fread (buffer, 1, 8, f); buffer [8] = '\0'; // number of samples in each data record
long numberOfSamplesInThisDataRecord = atol (buffer);
if (samplingFrequency == NUMundefined) {
numberOfSamplesPerDataRecord = numberOfSamplesInThisDataRecord;
samplingFrequency = numberOfSamplesInThisDataRecord / durationOfDataRecord;
}
if (numberOfSamplesInThisDataRecord / durationOfDataRecord != samplingFrequency)
Melder_throw (U"Number of samples per data record in channel ", channel,
U" (", numberOfSamplesInThisDataRecord,
U") doesn't match sampling frequency of channel 1 (", samplingFrequency, U").");
}
for (long channel = 1; channel <= numberOfChannels; channel ++) {
fread (buffer, 1, 32, f); buffer [32] = '\0'; // reserved
}
double duration = numberOfDataRecords * durationOfDataRecord;
autoEEG him = EEG_create (0, duration);
his numberOfChannels = numberOfChannels;
autoSound me = Sound_createSimple (numberOfChannels, duration, samplingFrequency);
Melder_assert (my nx == numberOfSamplesPerDataRecord * numberOfDataRecords);
autoNUMvector <unsigned char> dataBuffer (0L, 3 * numberOfSamplesPerDataRecord - 1);
for (long record = 1; record <= numberOfDataRecords; record ++) {
for (long channel = 1; channel <= numberOfChannels; channel ++) {
double factor = channel == numberOfChannels ? 1.0 : physicalMinimum [channel] / digitalMinimum [channel];
if (channel < numberOfChannels - EEG_getNumberOfExtraSensors (him.peek())) factor /= 1000000.0;
if (is24bit) {
fread (& dataBuffer [0], 3, numberOfSamplesPerDataRecord, f);
unsigned char *p = & dataBuffer [0];
for (long i = 1; i <= numberOfSamplesPerDataRecord; i ++) {
long sample = i + (record - 1) * numberOfSamplesPerDataRecord;
Melder_assert (sample <= my nx);
uint8_t lowByte = *p ++, midByte = *p ++, highByte = *p ++;
uint32_t externalValue = ((uint32_t) highByte << 16) | ((uint32_t) midByte << 8) | (uint32_t) lowByte;
if ((highByte & 128) != 0) // is the 24-bit sign bit on?
externalValue |= 0xFF000000; // extend negative sign to 32 bits
my z [channel] [sample] = (int32_t) externalValue * factor;
}
} else {
fread (& dataBuffer [0], 2, numberOfSamplesPerDataRecord, f);
unsigned char *p = & dataBuffer [0];
for (long i = 1; i <= numberOfSamplesPerDataRecord; i ++) {
long sample = i + (record - 1) * numberOfSamplesPerDataRecord;
Melder_assert (sample <= my nx);
uint8 lowByte = *p ++, highByte = *p ++;
uint16 externalValue = (uint16) ((uint16) highByte << 8) | (uint16) lowByte;
my z [channel] [sample] = (int16) externalValue * factor;
}
}
}
}
int numberOfStatusBits = 8;
for (long i = 1; i <= my nx; i ++) {
unsigned long value = (long) my z [numberOfChannels] [i];
if (value & 0x0000FF00) {
numberOfStatusBits = 16;
}
}
autoTextGrid thee;
if (hasLetters) {
thee = TextGrid_create (0, duration, U"Mark Trigger", U"Mark Trigger");
autoMelderString letters;
double time = NUMundefined;
for (long i = 1; i <= my nx; i ++) {
unsigned long value = (long) my z [numberOfChannels] [i];
for (int byte = 1; byte <= numberOfStatusBits / 8; byte ++) {
unsigned long mask = byte == 1 ? 0x000000ff : 0x0000ff00;
char32 kar = byte == 1 ? (value & mask) : (value & mask) >> 8;
if (kar != U'\0' && kar != 20) {
MelderString_appendCharacter (& letters, kar);
} else if (letters. string [0] != U'\0') {
if (letters. string [0] == U'+') {
if (NUMdefined (time)) {
try {
TextGrid_insertPoint (thee.peek(), 1, time, U"");
} catch (MelderError) {
Melder_throw (U"Did not insert empty mark (", letters. string, U") on Mark tier.");
}
time = NUMundefined; // defensive
}
time = Melder_atof (& letters. string [1]);
MelderString_empty (& letters);
} else {
if (! NUMdefined (time)) {
Melder_throw (U"Undefined time for label at sample ", i, U".");
}
try {
if (Melder_nequ (letters. string, U"Trigger-", 8)) {
try {
TextGrid_insertPoint (thee.peek(), 2, time, & letters. string [8]);
} catch (MelderError) {
Melder_clearError ();
trace (U"Duplicate trigger at ", time, U" seconds: ", & letters. string [8]);
}
} else {
TextGrid_insertPoint (thee.peek(), 1, time, & letters. string [0]);
}
} catch (MelderError) {
Melder_throw (U"Did not insert mark (", letters. string, U") on Trigger tier.");
}
time = NUMundefined; // crucial
MelderString_empty (& letters);
}
}
}
}
if (NUMdefined (time)) {
TextGrid_insertPoint (thee.peek(), 1, time, U"");
time = NUMundefined; // defensive
}
} else {
thee = TextGrid_create (0, duration,
numberOfStatusBits == 8 ? U"S1 S2 S3 S4 S5 S6 S7 S8" : U"S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16", U"");
for (int bit = 1; bit <= numberOfStatusBits; bit ++) {
unsigned long bitValue = 1 << (bit - 1);
IntervalTier tier = (IntervalTier) thy tiers -> item [bit];
for (long i = 1; i <= my nx; i ++) {
unsigned long previousValue = i == 1 ? 0 : (long) my z [numberOfChannels] [i - 1];
unsigned long thisValue = (long) my z [numberOfChannels] [i];
if ((thisValue & bitValue) != (previousValue & bitValue)) {
double time = i == 1 ? 0.0 : my x1 + (i - 1.5) * my dx;
if (time != 0.0)
TextGrid_insertBoundary (thee.peek(), bit, time);
if ((thisValue & bitValue) != 0)
TextGrid_setIntervalText (thee.peek(), bit, tier -> intervals -> size, U"1");
}
}
}
}
f.close (file);
his channelNames = channelNames.transfer();
his sound = me.move();
his textgrid = thee.move();
if (EEG_getNumberOfCapElectrodes (him.peek()) == 32) {
EEG_setChannelName (him.peek(), 1, U"Fp1");
EEG_setChannelName (him.peek(), 2, U"AF3");
EEG_setChannelName (him.peek(), 3, U"F7");
EEG_setChannelName (him.peek(), 4, U"F3");
EEG_setChannelName (him.peek(), 5, U"FC1");
EEG_setChannelName (him.peek(), 6, U"FC5");
EEG_setChannelName (him.peek(), 7, U"T7");
EEG_setChannelName (him.peek(), 8, U"C3");
EEG_setChannelName (him.peek(), 9, U"CP1");
EEG_setChannelName (him.peek(), 10, U"CP5");
EEG_setChannelName (him.peek(), 11, U"P7");
EEG_setChannelName (him.peek(), 12, U"P3");
EEG_setChannelName (him.peek(), 13, U"Pz");
EEG_setChannelName (him.peek(), 14, U"PO3");
EEG_setChannelName (him.peek(), 15, U"O1");
EEG_setChannelName (him.peek(), 16, U"Oz");
EEG_setChannelName (him.peek(), 17, U"O2");
EEG_setChannelName (him.peek(), 18, U"PO4");
EEG_setChannelName (him.peek(), 19, U"P4");
EEG_setChannelName (him.peek(), 20, U"P8");
EEG_setChannelName (him.peek(), 21, U"CP6");
EEG_setChannelName (him.peek(), 22, U"CP2");
EEG_setChannelName (him.peek(), 23, U"C4");
EEG_setChannelName (him.peek(), 24, U"T8");
EEG_setChannelName (him.peek(), 25, U"FC6");
EEG_setChannelName (him.peek(), 26, U"FC2");
EEG_setChannelName (him.peek(), 27, U"F4");
EEG_setChannelName (him.peek(), 28, U"F8");
EEG_setChannelName (him.peek(), 29, U"AF4");
EEG_setChannelName (him.peek(), 30, U"Fp2");
EEG_setChannelName (him.peek(), 31, U"Fz");
EEG_setChannelName (him.peek(), 32, U"Cz");
} else if (EEG_getNumberOfCapElectrodes (him.peek()) == 64) {
EEG_setChannelName (him.peek(), 1, U"Fp1");
EEG_setChannelName (him.peek(), 2, U"AF7");
EEG_setChannelName (him.peek(), 3, U"AF3");
EEG_setChannelName (him.peek(), 4, U"F1");
EEG_setChannelName (him.peek(), 5, U"F3");
EEG_setChannelName (him.peek(), 6, U"F5");
EEG_setChannelName (him.peek(), 7, U"F7");
EEG_setChannelName (him.peek(), 8, U"FT7");
EEG_setChannelName (him.peek(), 9, U"FC5");
EEG_setChannelName (him.peek(), 10, U"FC3");
EEG_setChannelName (him.peek(), 11, U"FC1");
EEG_setChannelName (him.peek(), 12, U"C1");
EEG_setChannelName (him.peek(), 13, U"C3");
EEG_setChannelName (him.peek(), 14, U"C5");
EEG_setChannelName (him.peek(), 15, U"T7");
EEG_setChannelName (him.peek(), 16, U"TP7");
EEG_setChannelName (him.peek(), 17, U"CP5");
EEG_setChannelName (him.peek(), 18, U"CP3");
EEG_setChannelName (him.peek(), 19, U"CP1");
EEG_setChannelName (him.peek(), 20, U"P1");
EEG_setChannelName (him.peek(), 21, U"P3");
EEG_setChannelName (him.peek(), 22, U"P5");
EEG_setChannelName (him.peek(), 23, U"P7");
EEG_setChannelName (him.peek(), 24, U"P9");
EEG_setChannelName (him.peek(), 25, U"PO7");
EEG_setChannelName (him.peek(), 26, U"PO3");
EEG_setChannelName (him.peek(), 27, U"O1");
EEG_setChannelName (him.peek(), 28, U"Iz");
EEG_setChannelName (him.peek(), 29, U"Oz");
EEG_setChannelName (him.peek(), 30, U"POz");
EEG_setChannelName (him.peek(), 31, U"Pz");
EEG_setChannelName (him.peek(), 32, U"CPz");
EEG_setChannelName (him.peek(), 33, U"Fpz");
EEG_setChannelName (him.peek(), 34, U"Fp2");
EEG_setChannelName (him.peek(), 35, U"AF8");
EEG_setChannelName (him.peek(), 36, U"AF4");
EEG_setChannelName (him.peek(), 37, U"AFz");
EEG_setChannelName (him.peek(), 38, U"Fz");
EEG_setChannelName (him.peek(), 39, U"F2");
EEG_setChannelName (him.peek(), 40, U"F4");
EEG_setChannelName (him.peek(), 41, U"F6");
EEG_setChannelName (him.peek(), 42, U"F8");
EEG_setChannelName (him.peek(), 43, U"FT8");
EEG_setChannelName (him.peek(), 44, U"FC6");
EEG_setChannelName (him.peek(), 45, U"FC4");
EEG_setChannelName (him.peek(), 46, U"FC2");
EEG_setChannelName (him.peek(), 47, U"FCz");
EEG_setChannelName (him.peek(), 48, U"Cz");
EEG_setChannelName (him.peek(), 49, U"C2");
EEG_setChannelName (him.peek(), 50, U"C4");
EEG_setChannelName (him.peek(), 51, U"C6");
EEG_setChannelName (him.peek(), 52, U"T8");
EEG_setChannelName (him.peek(), 53, U"TP8");
EEG_setChannelName (him.peek(), 54, U"CP6");
EEG_setChannelName (him.peek(), 55, U"CP4");
EEG_setChannelName (him.peek(), 56, U"CP2");
EEG_setChannelName (him.peek(), 57, U"P2");
EEG_setChannelName (him.peek(), 58, U"P4");
EEG_setChannelName (him.peek(), 59, U"P6");
EEG_setChannelName (him.peek(), 60, U"P8");
EEG_setChannelName (him.peek(), 61, U"P10");
EEG_setChannelName (him.peek(), 62, U"PO8");
EEG_setChannelName (him.peek(), 63, U"PO4");
EEG_setChannelName (him.peek(), 64, U"O2");
}
return him;
} catch (MelderError) {
autoSpectrogram Sound_and_Pitch_to_Spectrogram (Sound me, Pitch thee, double analysisWidth, double dt, double f1_hz, double fmax_hz, double df_hz, double relative_bw) {
try {
double t1, windowDuration = 2.0 * analysisWidth; /* gaussian window */
double nyquist = 0.5 / my dx, samplingFrequency = 2.0 * nyquist, fmin_hz = 0.0;
long numberOfFrames, f0_undefined = 0.0;
if (my xmin > thy xmin || my xmax > thy xmax) Melder_throw
(U"The domain of the Sound is not included in the domain of the Pitch.");
double f0_median = Pitch_getQuantile (thee, thy xmin, thy xmax, 0.5, kPitch_unit_HERTZ);
if (f0_median == NUMundefined || f0_median == 0.0) {
f0_median = 100.0;
Melder_warning (U"Pitch values undefined. Bandwith fixed to 100 Hz. ");
}
if (f1_hz <= 0.0) {
f1_hz = 100.0;
}
if (fmax_hz <= 0.0) {
fmax_hz = nyquist;
}
if (df_hz <= 0.0) {
df_hz = f0_median / 2.0;
}
if (relative_bw <= 0.0) {
relative_bw = 1.1;
}
fmax_hz = MIN (fmax_hz, nyquist);
long numberOfFilters = lround ( (fmax_hz - f1_hz) / df_hz);
Sampled_shortTermAnalysis (me, windowDuration, dt, &numberOfFrames, &t1);
autoSpectrogram him = Spectrogram_create (my xmin, my xmax, numberOfFrames, dt, t1, fmin_hz, fmax_hz, numberOfFilters, df_hz, f1_hz);
// Temporary objects
autoSound sframe = Sound_createSimple (1, windowDuration, samplingFrequency);
autoSound window = Sound_createGaussian (windowDuration, samplingFrequency);
autoMelderProgress progress (U"Sound & Pitch: To FormantFilter");
for (long iframe = 1; iframe <= numberOfFrames; iframe++) {
double t = Sampled_indexToX (him.get(), iframe);
double b, f0 = Pitch_getValueAtTime (thee, t, kPitch_unit_HERTZ, 0);
if (f0 == NUMundefined || f0 == 0.0) {
f0_undefined ++;
f0 = f0_median;
}
b = relative_bw * f0;
Sound_into_Sound (me, sframe.get(), t - windowDuration / 2.0);
Sounds_multiply (sframe.get(), window.get());
Sound_into_Spectrogram_frame (sframe.get(), him.get(), iframe, b);
if (iframe % 10 == 1) {
Melder_progress ( (double) iframe / numberOfFrames, U"Frame ", iframe, U" out of ",
numberOfFrames, U".");
}
}
_Spectrogram_windowCorrection (him.get(), window -> nx);
return him;
} catch (MelderError) {
Melder_throw (U"FormantFilter not created from Pitch & FormantFilter.");
}
}
BarkFilter Sound_to_BarkFilter (Sound me, double analysisWidth, double dt,
double f1_bark, double fmax_bark, double df_bark) {
try {
double t1, nyquist = 0.5 / my dx, samplingFrequency = 2 * nyquist;
double windowDuration = 2 * analysisWidth; /* gaussian window */
double zmax = NUMhertzToBark2 (nyquist);
double fmin_bark = 0;
long nt, frameErrorCount = 0;
// Check defaults.
if (f1_bark <= 0) {
f1_bark = 1;
}
if (fmax_bark <= 0) {
fmax_bark = zmax;
}
if (df_bark <= 0) {
df_bark = 1;
}
fmax_bark = MIN (fmax_bark, zmax);
long nf = floor ( (fmax_bark - f1_bark) / df_bark + 0.5);
if (nf <= 0) {
Melder_throw ("The combination of filter parameters is not valid.");
}
Sampled_shortTermAnalysis (me, windowDuration, dt, & nt, & t1);
autoSound sframe = Sound_createSimple (1, windowDuration, samplingFrequency);
autoSound window = Sound_createGaussian (windowDuration, samplingFrequency);
autoBarkFilter thee = BarkFilter_create (my xmin, my xmax, nt, dt, t1,
fmin_bark, fmax_bark, nf, df_bark, f1_bark);
autoMelderProgress progess (L"BarkFilter analysis");
for (long i = 1; i <= nt; i++) {
double t = Sampled_indexToX (thee.peek(), i);
Sound_into_Sound (me, sframe.peek(), t - windowDuration / 2);
Sounds_multiply (sframe.peek(), window.peek());
if (! Sound_into_BarkFilter_frame (sframe.peek(), thee.peek(), i)) {
frameErrorCount++;
}
if ( (i % 10) == 1) {
Melder_progress ( (double) i / nt, L"BarkFilter analysis: frame ",
Melder_integer (i), L" from ", Melder_integer (nt), L"."); therror
}
}
if (frameErrorCount > 0) {
Melder_warning (L"Analysis results of ", Melder_integer (frameErrorCount), L" frame(s) out of ",
Melder_integer (nt), L" will be suspect.");
}
double ref = FilterBank_DBREF * gaussian_window_squared_correction (window -> nx);
NUMdmatrix_to_dBs (thy z, 1, thy ny, 1, thy nx, ref, FilterBank_DBFAC, FilterBank_DBFLOOR);
return thee.transfer();
} catch (MelderError) {
Sound Sound_recordFixedTime (int inputSource, double gain, double balance, double sampleRate, double duration) {
bool inputUsesPortAudio = MelderAudio_getInputUsesPortAudio ();
PaStream *portaudioStream = NULL;
#if defined (macintosh)
long refNum;
#elif defined (_WIN32)
HWAVEIN hWaveIn = 0;
#else
int fd = -1; /* Other systems use stream I/O with a file descriptor. */
int fd_mixer = -1;
#endif
try {
long numberOfSamples, i;
int mulaw = FALSE;
int can16bit = TRUE;
int fakeMonoByStereo = FALSE; /* Will be set to TRUE for systems (like MacOS X) that do not allow direct mono recording. */
/* Declare system-dependent data structures. */
static bool paInitialized = false;
volatile struct Sound_recordFixedTime_Info info = { 0 };
PaStreamParameters streamParameters = { 0 };
#if defined (macintosh)
#elif defined (_WIN32)
WAVEFORMATEX waveFormat;
WAVEHDR waveHeader;
MMRESULT err;
(void) inputSource;
(void) gain;
(void) balance;
#elif defined (linux)
#define min(a,b) a > b ? b : a
int dev_mask;
int fd_mixer = -1;
int val;
#endif
/* Check representation of shorts. */
if (sizeof (short) != 2)
Melder_throw ("Cannot record a sound on this computer.");
/* Check sampling frequency. */
bool supportsSamplingFrequency = true;
if (inputUsesPortAudio) {
#if defined (macintosh)
if (sampleRate != 44100 && sampleRate != 48000 && sampleRate != 96000) supportsSamplingFrequency = false;
#endif
} else {
#if defined (macintosh)
if (sampleRate != 44100) supportsSamplingFrequency = false;
#elif defined (linux)
if (sampleRate != 8000 && sampleRate != 11025 &&
sampleRate != 16000 && sampleRate != 22050 &&
sampleRate != 32000 && sampleRate != 44100 &&
sampleRate != 48000) supportsSamplingFrequency = false;
#elif defined (_WIN32)
if (sampleRate != 8000 && sampleRate != 11025 &&
sampleRate != 16000 && sampleRate != 22050 &&
sampleRate != 32000 && sampleRate != 44100 &&
sampleRate != 48000 && sampleRate != 96000) supportsSamplingFrequency = false;
#endif
}
if (! supportsSamplingFrequency)
Melder_throw ("Your audio hardware does not support a sampling frequency of ", sampleRate, " Hz.");
/*
* Open phase 1.
* On some systems, the info is filled in before the audio port is opened.
* On other systems, the info is filled in after the port is opened.
*/
if (inputUsesPortAudio) {
if (! paInitialized) {
PaError err = Pa_Initialize ();
if (err)
Melder_throw ("Pa_Initialize: ", Pa_GetErrorText (err));
paInitialized = true;
}
} else {
#if defined (macintosh)
#elif defined (_WIN32)
#else
/* We must open the port now, because we use an ioctl to set the info to an open port. */
fd = open (DEV_AUDIO, O_RDONLY);
if (fd == -1) {
if (errno == EBUSY)
Melder_throw ("Audio device in use by another program.");
else
#ifdef linux
Melder_throw ("Cannot open audio device.\nPlease switch on PortAudio in the Sound Recording Preferences.");
#else
Melder_throw ("Cannot open audio device.");
#endif
}
/* The device immediately started recording into its buffer, but probably at the wrong rate etc. */
/* Pause and flush this rubbish. */
#if defined (linux)
ioctl (fd, SNDCTL_DSP_RESET, NULL);
#endif
#endif
}
/* Set the input source; the default is the microphone. */
if (inputUsesPortAudio) {
if (inputSource < 1 || inputSource > Pa_GetDeviceCount ())
Melder_throw ("Unknown device #", inputSource, ".");
streamParameters. device = inputSource - 1;
} else {
#if defined (macintosh)
#elif defined (linux)
fd_mixer = open ("/dev/mixer", O_WRONLY);
if (fd_mixer == -1)
Melder_throw ("Cannot open /dev/mixer.");
dev_mask = inputSource == 1 ? SOUND_MASK_MIC : SOUND_MASK_LINE;
if (ioctl (fd_mixer, SOUND_MIXER_WRITE_RECSRC, & dev_mask) == -1)
Melder_throw ("Cannot set recording device in mixer");
#endif
}
/* Set gain and balance. */
if (inputUsesPortAudio) {
/* Taken from Audio Control Panel. */
} else {
#if defined (macintosh) || defined (_WIN32)
/* Taken from Audio Control Panel. */
#elif defined (linux)
val = (gain <= 0.0 ? 0 : gain >= 1.0 ? 100 : floor (gain * 100 + 0.5));
balance = balance <= 0 ? 0 : balance >= 1 ? 1 : balance;
if (balance >= 0.5) {
val = (int)(((int)(val*balance/(1-balance)) << 8) | val);
} else {
val = (int)(val | ((int)(val*(1-balance)/balance) << 8));
}
val = (int)((min(2-2*balance,1))*val) | ((int)((min(2*balance,1))*val) << 8);
if (inputSource == 1) {
/* MIC */
if (ioctl (fd_mixer, MIXER_WRITE (SOUND_MIXER_MIC), & val) == -1)
Melder_throw ("Cannot set gain and balance.");
} else {
/* LINE */
if (ioctl (fd_mixer, MIXER_WRITE (SOUND_MIXER_LINE), & val) == -1)
Melder_throw ("Cannot set gain and balance.");
}
close (fd_mixer);
fd_mixer = -1;
#endif
}
/* Set the sampling frequency. */
if (inputUsesPortAudio) {
// Set while opening.
} else {
#if defined (macintosh)
#elif defined (linux)
int sampleRate_int = (int) sampleRate;
if (ioctl (fd, SNDCTL_DSP_SPEED, & sampleRate_int) == -1)
Melder_throw ("Cannot set sampling frequency to ", sampleRate, " Hz.");
#elif defined (_WIN32)
waveFormat. nSamplesPerSec = sampleRate;
#endif
}
/* Set the number of channels to 1 (mono), if possible. */
if (inputUsesPortAudio) {
streamParameters. channelCount = 1;
} else {
#if defined (macintosh)
#elif defined (linux)
val = 1;
if (ioctl (fd, SNDCTL_DSP_CHANNELS, & val) == -1)
Melder_throw ("Cannot set to mono.");
#elif defined (_WIN32)
waveFormat. nChannels = 1;
#endif
}
/* Set the encoding to 16-bit linear (or to 8-bit linear, if 16-bit is not available). */
if (inputUsesPortAudio) {
streamParameters. sampleFormat = paInt16;
} else {
#if defined (macintosh)
#elif defined (linux)
#if __BYTE_ORDER == __BIG_ENDIAN
val = AFMT_S16_BE;
#else
val = AFMT_S16_LE;
#endif
if (ioctl (fd, SNDCTL_DSP_SETFMT, & val) == -1)
Melder_throw ("Cannot set 16-bit linear.");
#elif defined (_WIN32)
waveFormat. wFormatTag = WAVE_FORMAT_PCM;
waveFormat. wBitsPerSample = 16;
waveFormat. nBlockAlign = waveFormat. nChannels * waveFormat. wBitsPerSample / 8;
waveFormat. nAvgBytesPerSec = waveFormat. nBlockAlign * waveFormat. nSamplesPerSec;
#endif
}
/* Create a buffer for recording, and the resulting sound. */
numberOfSamples = floor (sampleRate * duration + 0.5);
if (numberOfSamples < 1)
Melder_throw ("Duration too short.");
autoNUMvector <short> buffer (1, numberOfSamples * (fakeMonoByStereo ? 2 : 1));
autoSound me = Sound_createSimple (1, numberOfSamples / sampleRate, sampleRate); // STEREO BUG
Melder_assert (my nx == numberOfSamples);
/*
* Open phase 2.
* This starts recording now.
*/
if (inputUsesPortAudio) {
streamParameters. suggestedLatency = Pa_GetDeviceInfo (inputSource - 1) -> defaultLowInputLatency;
#if defined (macintosh)
PaMacCoreStreamInfo macCoreStreamInfo = { 0 };
macCoreStreamInfo. size = sizeof (PaMacCoreStreamInfo);
macCoreStreamInfo. hostApiType = paCoreAudio;
macCoreStreamInfo. version = 0x01;
macCoreStreamInfo. flags = paMacCoreChangeDeviceParameters | paMacCoreFailIfConversionRequired;
streamParameters. hostApiSpecificStreamInfo = & macCoreStreamInfo;
#endif
info. numberOfSamples = numberOfSamples;
info. numberOfSamplesRead = 0;
info. buffer = buffer.peek();
PaError err = Pa_OpenStream (& portaudioStream, & streamParameters, NULL,
sampleRate, 0, paNoFlag, portaudioStreamCallback, (void *) & info);
if (err)
Melder_throw ("open ", Pa_GetErrorText (err));
Pa_StartStream (portaudioStream);
if (err)
Melder_throw ("start ", Pa_GetErrorText (err));
} else {
#if defined (macintosh)
#elif defined (_WIN32)
waveFormat. cbSize = 0;
err = waveInOpen (& hWaveIn, WAVE_MAPPER, & waveFormat, 0, 0, CALLBACK_NULL);
if (err != MMSYSERR_NOERROR)
Melder_throw ("Error ", err, " while opening.");
#endif
}
/* Read the sound into the buffer. */
if (inputUsesPortAudio) {
// The callback will do this. Just wait.
while (/*getNumberOfSamplesRead (& info)*/ info. numberOfSamplesRead < numberOfSamples) {
//Pa_Sleep (1);
//Melder_casual ("filled %ld/%ld", getNumberOfSamplesRead (& info), numberOfSamples);
}
} else {
#if defined (macintosh)
#elif defined (_WIN32)
waveHeader. dwFlags = 0;
waveHeader. lpData = (char *) & buffer [1];
waveHeader. dwBufferLength = numberOfSamples * 2;
waveHeader. dwLoops = 0;
waveHeader. lpNext = NULL;
waveHeader. reserved = 0;
err = waveInPrepareHeader (hWaveIn, & waveHeader, sizeof (WAVEHDR));
if (err != MMSYSERR_NOERROR)
Melder_throw ("Error ", err, " while preparing header.");
err = waveInAddBuffer (hWaveIn, & waveHeader, sizeof (WAVEHDR));
if (err != MMSYSERR_NOERROR)
Melder_throw ("Error ", err, " while listening.");
err = waveInStart (hWaveIn);
if (err != MMSYSERR_NOERROR)
Melder_throw ("Error ", err, " while starting.");
while (! (waveHeader. dwFlags & WHDR_DONE)) { Pa_Sleep (1); }
err = waveInUnprepareHeader (hWaveIn, & waveHeader, sizeof (WAVEHDR));
if (err != MMSYSERR_NOERROR)
Melder_throw ("Error ", err, " while unpreparing header.");
#else
if (mulaw)
read (fd, (char *) & buffer [1], numberOfSamples);
else {
long bytesLeft = 2 * numberOfSamples, dbytes, bytesRead = 0;
while (bytesLeft) {
//Melder_casual ("Reading %ld bytes", bytesLeft > 4000 ? 4000 : bytesLeft);
dbytes = read (fd, & ((char *) buffer.peek()) [2 + bytesRead], bytesLeft > 4000 ? 4000 : bytesLeft);
//Melder_casual("Read %ld bytes", dbytes);
if (dbytes <= 0) break;
bytesLeft -= dbytes;
bytesRead += dbytes;
};
}
#endif
}
/* Copy the buffered data to the sound object, and discard the buffer. */
if (fakeMonoByStereo)
for (i = 1; i <= numberOfSamples; i ++)
my z [1] [i] = ((long) buffer [i + i - 1] + buffer [i + i]) * (1.0 / 65536);
else if (mulaw)
for (i = 1; i <= numberOfSamples; i ++)
my z [1] [i] = ulaw2linear [((unsigned char *) buffer.peek()) [i]] * (1.0 / 32768);
else if (can16bit)
for (i = 1; i <= numberOfSamples; i ++)
my z [1] [i] = buffer [i] * (1.0 / 32768);
else
for (i = 1; i <= numberOfSamples; i ++)
my z [1] [i] = ((int) ((unsigned char *) buffer.peek()) [i + 1] - 128) * (1.0 / 128);
/* Close the audio device. */
if (inputUsesPortAudio) {
Pa_StopStream (portaudioStream);
Pa_CloseStream (portaudioStream);
} else {
#if defined (macintosh)
#elif defined (_WIN32)
err = waveInClose (hWaveIn);
if (err != MMSYSERR_NOERROR)
Melder_throw ("Error ", err, " while closing.");
#else
close (fd);
#endif
}
/* Hand the resulting sound to the caller. */
return me.transfer();
} catch (MelderError) {
if (inputUsesPortAudio) {
if (portaudioStream) Pa_StopStream (portaudioStream);
if (portaudioStream) Pa_CloseStream (portaudioStream);
} else {
#if defined (macintosh)
#elif defined (_WIN32)
if (hWaveIn != 0) waveInClose (hWaveIn);
#else
if (fd_mixer != -1) close (fd_mixer);
if (fd != -1) close (fd);
#endif
}
Melder_throw ("Sound not recorded.");
}
}
SPINET Sound_to_SPINET (Sound me, double timeStep, double windowDuration,
double minimumFrequencyHz, double maximumFrequencyHz, long nFilters,
double excitationErbProportion, double inhibitionErbProportion) {
try {
double firstTime, b = 1.02, samplingFrequency = 1 / my dx;
if (timeStep < my dx) {
timeStep = my dx;
}
if (maximumFrequencyHz > samplingFrequency / 2) {
maximumFrequencyHz = samplingFrequency / 2;
}
long numberOfFrames;
Sampled_shortTermAnalysis (me, windowDuration, timeStep, &numberOfFrames, &firstTime);
autoSPINET thee = SPINET_create (my xmin, my xmax, numberOfFrames, timeStep, firstTime,
minimumFrequencyHz, maximumFrequencyHz, nFilters, excitationErbProportion, inhibitionErbProportion);
autoSound window = Sound_createGaussian (windowDuration, samplingFrequency);
autoSound frame = Sound_createSimple (1, windowDuration, samplingFrequency);
autoNUMvector<double> f (1, nFilters);
autoNUMvector<double> bw (1, nFilters);
autoNUMvector<double> aex (1, nFilters);
autoNUMvector<double> ain (1, nFilters);
// Cochlear filterbank: gammatone
for (long i = 1; i <= nFilters; i++) {
f[i] = NUMerbToHertz (thy y1 + (i - 1) * thy dy);
bw[i] = 2 * NUMpi * b * (f[i] * (6.23e-6 * f[i] + 93.39e-3) + 28.52);
}
autoMelderProgress progress (L"SPINET analysis");
for (long i = 1; i <= nFilters; i++) {
double bb = (f[i] / 1000) * exp (- f[i] / 1000); // outer & middle ear and phase locking
double tgammaMax = (thy gamma - 1) / bw[i]; // Time where gammafunction envelope has maximum
double gammaMaxAmplitude = pow ( (thy gamma - 1) / (NUMe * bw[i]), (thy gamma - 1)); // tgammaMax
double timeCorrection = tgammaMax - windowDuration / 2;
autoSound gammaTone = Sound_createGammaTone (0, 0.1, samplingFrequency,
thy gamma, b, f[i], 0, 0, 0);
autoSound filtered = Sounds_convolve (me, gammaTone.peek(), kSounds_convolve_scaling_SUM, kSounds_convolve_signalOutsideTimeDomain_ZERO);
// To energy measure: weigh with broad-band transfer function
for (long j = 1; j <= numberOfFrames; j++) {
Sound_into_Sound (filtered.peek(), frame.peek(), Sampled_indexToX (thee.peek(), j) + timeCorrection);
Sounds_multiply (frame.peek(), window.peek());
thy y[i][j] = Sound_power (frame.peek()) * bb / gammaMaxAmplitude;
}
Melder_progress ( (double) i / nFilters, L"SPINET: filter ", Melder_integer (i), L" from ",
Melder_integer (nFilters), L".");
}
// Excitatory and inhibitory area functions
for (long i = 1; i <= nFilters; i++) {
for (long k = 1; k <= nFilters; k++) {
double fr = (f[k] - f[i]) / bw[i];
aex[i] += fgamma (fr / thy excitationErbProportion, thy gamma);
ain[i] += fgamma (fr / thy inhibitionErbProportion, thy gamma);
}
}
// On-center off-surround interactions
for (long j = 1; j <= numberOfFrames; j++)
for (long i = 1; i <= nFilters; i++) {
double a = 0;
for (long k = 1; k <= nFilters; k++) {
double fr = (f[k] - f[i]) / bw[i];
double hexsq = fgamma (fr / thy excitationErbProportion, thy gamma);
double hinsq = fgamma (fr / thy inhibitionErbProportion, thy gamma);
a += thy y[k][j] * (hexsq / aex[i] - hinsq / ain[i]);
}
thy s[i][j] = a > 0 ? a : 0;
}
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, ": no SPINET created.");
}
}
LPC LPC_and_Sound_to_LPC_robust (LPC thee, Sound me, double analysisWidth, double preEmphasisFrequency, double k,
int itermax, double tol, int wantlocation) {
struct huber_struct struct_huber = { 0 };
try {
double t1, samplingFrequency = 1.0 / my dx, tol_svd = 0.000001;
double location = 0, windowDuration = 2 * analysisWidth; /* Gaussian window */
long nFrames, frameErrorCount = 0, iter = 0;
long p = thy maxnCoefficients;
if (my xmin != thy xmin || my xmax != thy xmax) {
Melder_throw ("Time domains differ.");
}
if (my dx != thy samplingPeriod) {
Melder_throw ("Sampling intervals differ.");
}
if (floor (windowDuration / my dx) < p + 1) {
Melder_throw ("Analysis window too short.");
}
Sampled_shortTermAnalysis (me, windowDuration, thy dx, & nFrames, & t1);
if (nFrames != thy nx || t1 != thy x1) {
Melder_throw ("Incorrect retrieved analysis width");
}
autoSound sound = Data_copy (me);
autoSound sframe = Sound_createSimple (1, windowDuration, samplingFrequency);
autoSound window = Sound_createGaussian (windowDuration, samplingFrequency);
autoLPC him = Data_copy (thee);
huber_struct_init (&struct_huber, windowDuration, p, samplingFrequency, location, wantlocation);
struct_huber.k = k;
struct_huber.tol = tol;
struct_huber.tol_svd = tol_svd;
struct_huber.itermax = itermax;
autoMelderProgress progess (L"LPC analysis");
Sound_preEmphasis (sound.peek(), preEmphasisFrequency);
for (long i = 1; i <= nFrames; i++) {
LPC_Frame lpc = (LPC_Frame) & thy d_frames[i];
LPC_Frame lpcto = (LPC_Frame) & his d_frames[i];
double t = Sampled_indexToX (thee, i);
Sound_into_Sound (sound.peek(), sframe.peek(), t - windowDuration / 2);
Vector_subtractMean (sframe.peek());
Sounds_multiply (sframe.peek(), window.peek());
try {
LPC_Frames_and_Sound_huber (lpc, sframe.peek(), lpcto, & struct_huber);
} catch (MelderError) {
frameErrorCount++;
}
iter += struct_huber.iter;
if ( (i % 10) == 1) {
Melder_progress ( (double) i / nFrames, L"LPC analysis of frame ",
Melder_integer (i), L" out of ", Melder_integer (nFrames), L".");
}
}
if (frameErrorCount) Melder_warning (L"Results of ", Melder_integer (frameErrorCount),
L" frame(s) out of ", Melder_integer (nFrames), L" could not be optimised.");
MelderInfo_writeLine4 (L"Number of iterations: ", Melder_integer (iter),
L"\n Average per frame: ", Melder_double (((double) iter) / nFrames));
huber_struct_destroy (&struct_huber);
return him.transfer();
} catch (MelderError) {
huber_struct_destroy (&struct_huber);
Melder_throw (me, ": no robust LPC created.");
}
}