DATASET *make_data(int points, double noise)
{
int i;
double x, y;
SERIES *ser;
DATASET *data;
ser = series_create();
ser->y_width = ser->x_delta = ser->y_delta = ser->offset = 1;
ser->x_width = 2; ser->step = 3;
/* Fill up a SERIES with 'points' patterns that consists of a single
* input, and 'nout' outputs. Each output is a successive delay from
* the logistic map.
*/
for(i = 0; i < points; i++) {
x = random_range(-1, 1);
y = random_range(-1, 1);
series_append_val(ser, x);
series_append_val(ser, y);
series_append_val(ser, sin(5 * x * y) + y);
}
data = dataset_create(&dsm_series_method, ser);
return(data);
}
fextract_series_t *getOnlyCoG(dsp_sample_t *signal, int nframes) {
int n_samples, offset=0, frames=0, i;
dsp_sample_t s[MAX_SAMPLES], last_s[MAX_SAMPLES];
dsp_sample_t *frame, *last_frame, *__tmp;
mx_real_t *window;
mx_complex_t *z;
fextract_series_t *serie=NULL;
serie=series_create(MAX_SERIES);
frame = s;
last_frame = last_s;
window = (mx_real_t *) rs_malloc(sizeof(mx_real_t) * SP_FRAME_LEN, "w");
z = (mx_complex_t *) rs_malloc(sizeof(mx_complex_t) * SP_FRAME_LEN, "z");
n_samples = next_mfcc_frame(frame, NULL, SP_FRAME_LEN, SP_FRAME_RATE, signal,nframes);
offset += SP_FRAME_LEN;
while (n_samples > 0) {
if (n_samples < SP_FRAME_LEN)
break;
frames++;
// FFT: ev. mit prae-emphase und zero-padding
/* ... Hamming-Fenstern ... */
dsp_window_hamming(window, frame, SP_FRAME_LEN);
for (i = 0; i < SP_FRAME_LEN; i++) {
mx_re(z[i]) = window[i];
mx_im(z[i]) = 0.0;
}
/* FFT */
dsp_xfft(z, SP_FRAME_LEN, 0);
series_add(serie,get_CoG(z,SP_FRAME_LEN));
/* ... und naechsten Frame einlesen */
__tmp = last_frame;
last_frame = frame;
frame = __tmp;
n_samples = next_mfcc_frame(frame, last_frame, SP_FRAME_LEN, SP_FRAME_RATE, signal+offset,nframes-offset);
offset += SP_FRAME_RATE;
}
rs_free(window);
rs_free(z);
return serie;
}
int getVoicingFeatures(mx_real_t *features, mx_real_t *pitch, int nframes) {
int i, voiced_length=0, unvoiced_length=0, fcount=0;
fextract_series_t *vseries = series_create(MAX_SERIES);
fextract_series_t *useries = series_create(MAX_SERIES);
if (nframes <=0) {
for (i=0;i<N_VOICING_FEATURES;i++)
features[fcount++]=-1;
return fcount;
}
for (i=0;i<nframes-1;i++) {
if (pitch[i]!=0)
voiced_length++;
else
unvoiced_length++;
if (i>0 && pitch[i-1] !=0 && pitch[i]==0) {
series_add(vseries,voiced_length);
voiced_length=0;
}
if (i>0 && pitch[i-1] ==0 && pitch[i]!=0) {
series_add(useries,unvoiced_length);
unvoiced_length=0;
}
}
if (pitch[nframes-1]!=0)
series_add(vseries,++voiced_length);
else
series_add(useries,++unvoiced_length);
fcount+=getStatistics(features+fcount,vseries->series,vseries->nSeries);
fcount+=getStatistics(features+fcount,useries->series,useries->nSeries);
features[fcount++]=1.0*vseries->nSeries/nframes;
series_destroy(vseries);
series_destroy(useries);
return fcount;
}
fextract_series_t* getPeaks (fextract_series_t *pulses, mx_real_t *signal, int nsamples, mx_real_t pmin, mx_real_t pmax, mx_real_t maximumPeriodFactor) {
int i, imin=0, imax=pulses->nSeries-1, numberOfPeaks=pulses->nSeries;
fextract_series_t *peaks;
if (numberOfPeaks < 3)
return NULL;
peaks = series_create(MAX_SERIES);
for (i = imin + 1; i < imax; i ++) {
mx_real_t p1 = pulses->series [i] - pulses->series [i - 1];
mx_real_t p2 = pulses->series [i + 1] - pulses->series [i];
mx_real_t intervalFactor = p1 > p2 ? p1 / p2 : p2 / p1;
if (pmin == pmax || (p1 >= pmin && p1 <= pmax && p2 >= pmin && p2 <= pmax && intervalFactor <= maximumPeriodFactor)) {
mx_real_t peak = getHannWindowedRms (signal, nsamples, pulses->series [i], 0.2 * p1, 0.2 * p2);
if (peak != MX_REAL_MAX && peak > 0.0) {
series_add(peaks,pulses->series[i]);
series_add(peaks,peak);
}
}
}
return peaks;
}
DATASET *make_data(int points)
{
int i;
double x1, x2;
SERIES *ser;
DATASET *data;
ser = series_create();
ser->x_width = ser->y_width = 2;
ser->x_delta = ser->y_delta = ser->offset = 1;
ser->step = 4;
for(i = 0; i < points; i++) {
x1 = random_range(-1, 1);
x2 = random_range(-1, 1);
series_append_val(ser, x1);
series_append_val(ser, x2);
series_append_val(ser, f1(x1, x2));
series_append_val(ser, f2(x1, x2));
}
data = dataset_create(&dsm_series_method, ser);
return(data);
}
int main(int argc, char **argv)
{
FILE *fp;
SERIES *ser;
double *x, *y;
unsigned i, j, sz;
unsigned vdx[XSZ - 1] = { 1, 2, 4, 8, 16 };
unsigned vdy[YSZ - 1] = { 3, 6 };
ser = series_create();
for(i = 0; i < SZ; i++)
series_append_val(ser, i);
ser->x_width = XSZ;
ser->y_width = YSZ;
ser->x_delta = 2;
ser->y_delta = 1;
ser->step = 1;
ser->offset = 2;
series_reinitiate(ser);
fp = fopen("tseries-xy.out", "w");
sz = series_get_num_pat(ser);
for(i = 0; i < sz; i++) {
x = series_get_x_pat(ser, i);
for(j = 0; j < XSZ; j++)
fprintf(fp, j < (XSZ - 1) ? "%d\t" : "%d\t\t", (int)x[j]);
y = series_get_y_pat(ser, i);
for(j = 0; j < YSZ; j++)
fprintf(fp, j < (YSZ - 1) ? "%d\t" : "%d\n", (int)y[j]);
}
fclose(fp);
ser->var_x_deltas = vdx;
ser->var_y_deltas = NULL;
series_reinitiate(ser);
fp = fopen("tseries-Xy.out", "w");
sz = series_get_num_pat(ser);
for(i = 0; i < sz; i++) {
x = series_get_x_pat(ser, i);
for(j = 0; j < XSZ; j++)
fprintf(fp, j < (XSZ - 1) ? "%d\t" : "%d\t\t", (int)x[j]);
y = series_get_y_pat(ser, i);
for(j = 0; j < YSZ; j++)
fprintf(fp, j < (YSZ - 1) ? "%d\t" : "%d\n", (int)y[j]);
}
fclose(fp);
ser->var_x_deltas = NULL;
ser->var_y_deltas = vdy;
series_reinitiate(ser);
fp = fopen("tseries-xY.out", "w");
sz = series_get_num_pat(ser);
for(i = 0; i < sz; i++) {
x = series_get_x_pat(ser, i);
for(j = 0; j < XSZ; j++)
fprintf(fp, j < (XSZ - 1) ? "%d\t" : "%d\t\t", (int)x[j]);
y = series_get_y_pat(ser, i);
for(j = 0; j < YSZ; j++)
fprintf(fp, j < (YSZ - 1) ? "%d\t" : "%d\n", (int)y[j]);
}
fclose(fp);
ser->var_x_deltas = vdx;
ser->var_y_deltas = vdy;
series_reinitiate(ser);
fp = fopen("tseries-XY.out", "w");
sz = series_get_num_pat(ser);
for(i = 0; i < sz; i++) {
x = series_get_x_pat(ser, i);
for(j = 0; j < XSZ; j++)
fprintf(fp, j < (XSZ - 1) ? "%d\t" : "%d\t\t", (int)x[j]);
y = series_get_y_pat(ser, i);
for(j = 0; j < YSZ; j++)
fprintf(fp, j < (YSZ - 1) ? "%d\t" : "%d\n", (int)y[j]);
}
fclose(fp);
exit(0);
}
fextract_series_t *pulses_calc(pitch_t *p, int nframes, mx_real_t *signal, int nsamples){
int i, x=0;
mx_real_t globalPeak=signal[0];
mx_real_t t = 0;
mx_real_t peak;
double addedRight = -1e300;
fextract_series_t *pulses = series_create(MAX_SERIES);
for (i=1;i<nsamples;i++) {
mx_real_t this = fabs(signal[i]);
if (this > globalPeak)
globalPeak=this;
}
/*
* Cycle over all voiced intervals.
*/
for (;;) {
mx_real_t tleft, tright, tmiddle, f0middle, tmax, tsave;
x++;
if (! getVoicedInterval (p, nframes, t, & tleft, & tright))
break;
/*
* Go to the middle of the voice stretch.
*/
tmiddle = (tleft + tright) / 2;
f0middle = getPitchValueAt(p,nframes,tmiddle);
/*
* Our first point is near this middle.
*/
tmax = findExtremum (signal, nsamples, tmiddle - 0.5 / f0middle, tmiddle + 0.5 / f0middle);
series_add(pulses,tmax);
//find pulses backward
tsave = tmax;
for (;;) {
mx_real_t f0 , correlation, last_tmax;
f0 = getPitchValueAt(p,nframes,tmax);
if (f0==-1)
break;
last_tmax=tmax;
correlation = findMaximumCorrelation (signal, nsamples, tmax, 1.0 / f0, tmax - 1.25 / f0, tmax - 0.8 / f0, & tmax, & peak);
if (correlation == -1 || tmax >last_tmax) /*break*/
tmax -= 1.0 / f0; /* This one period will drop out. */
if (tmax < tleft) {
if (correlation > 0.7 && peak > 0.023333 * globalPeak && tmax - addedRight > 0.8 / f0) {
series_add(pulses,tmax);
}
break;
}
if (correlation > 0.3 && (peak == 0.0 || peak > 0.01 * globalPeak)) {
if (tmax - addedRight > 0.8 / f0) { /* Do not fill in a short originally unvoiced interval twice. */
series_add(pulses,tmax);
}
}
}
//find pulses forward
tmax = tsave;
for (;;) {
mx_real_t f0 , correlation, last_tmax;
f0 = getPitchValueAt(p,nframes,tmax);
if (f0==-1)
break;
last_tmax=tmax;
correlation = findMaximumCorrelation (signal, nsamples, tmax, 1.0 / f0, tmax + 0.8 / f0, tmax + 1.25 / f0, & tmax, & peak);
if (correlation == -1 || tmax <last_tmax) /*break*/
tmax += 1.0 / f0;
if (tmax > tright) {
if (correlation > 0.7 && peak > 0.023333 * globalPeak) {
series_add(pulses,tmax);
addedRight = tmax;
}
break;
}
if (correlation > 0.3 && (peak == 0.0 || peak > 0.01 * globalPeak)) {
series_add(pulses,tmax);
addedRight = tmax;
}
}
t = tright;
}
qsort(pulses->series,pulses->nSeries,sizeof(mx_real_t),_cmp_mx_real);
return pulses;
}
int _fextract_calc(fextract_t *fex, mx_real_t *features, dsp_sample_t *signal, char *pfile) {
mx_real_t *pitch=NULL, *hnr=NULL;
mx_real_t **_mfcc_series, **_energy_series;
mx_real_t overall_max=0, overall_max_ind=0;
int nframes=0, i, j, n_voiced_frames, max;
int last_max=0, last_min=0, ser=0, max_sum=0, en_mean=0, fcount=0;
fextract_series_t * serie=NULL;
if (fex->frame_len==0) {
rs_warning("Cannot compute features for segment of size 0!");
return -1;
}
fex->n_features=V1_N_FEATURES;
_mfcc_series= (mx_real_t **) rs_malloc(MFCC_FEATURES * sizeof(mx_real_t *), "mfcc feature series");
for (i=0;i<MFCC_FEATURES;i++)
_mfcc_series[i]= (mx_real_t *) rs_malloc(MAX_FRAMES * sizeof(mx_real_t), "mfcc features");
_energy_series= (mx_real_t **) rs_malloc(ENERGY_FEATURES * sizeof(mx_real_t *), "energy feature series");
for (i=0;i<3;i++)
_energy_series[i]= (mx_real_t *) rs_malloc(MAX_FRAMES * sizeof(mx_real_t), "energy features");
if (pfile) {
FILE *pfile_fp= fopen(pfile,"r");
if (!pfile_fp)
rs_warning("pitch file %s not found!",pfile);
/* read external pitch values */
pitch = (mx_real_t *) rs_malloc(MAX_FRAMES * sizeof(mx_real_t),"external pitch values");
j=MAX_FRAMES;
i=0;
while (fscanf(pfile_fp,"%g",&pitch[i])==1) {
if (j == i+1) {
j+=MAX_FRAMES;
pitch = (mx_real_t *) rs_realloc(pitch,j*sizeof(mx_real_t),"external pitch values");
}
i++;
}
nframes=i;
fclose(pfile_fp);
}
else {
pitch= pitch_calc (fex->pitch,signal,fex->frame_len);
nframes=fex->pitch->nframes;
}
if (!pitch) {
for (;fcount<STATS*9+7;fcount++)
features[fcount]=-1.0;
n_voiced_frames=-1.0;
}
else {
const int series_size=9;
// @begin_change_johannes
//fextract_series_t *serien[series_size];
fextract_series_t *serien[9];
// @end_change_johannes
// printRawXML(pitch,nframes,"pitch");
pitch_frame_candidates_destroy(fex->pitch);
for (i=0;i<series_size;i++)
serien[i]=series_create(MAX_SERIES);
if (pitch[0]!=0)
series_add(serien[0],pitch[0]);
for (i=1;i<nframes-1;i++) {
if (pitch[i]!=0)
series_add(serien[0],pitch[i]);
if (pitch[i-1]<pitch[i] && pitch[i]>pitch[i+1]) {
series_add(serien[1],pitch[i]);
if (pitch[i] > overall_max) {
overall_max=pitch[i];
overall_max_ind=i;
}
series_add(serien[2],i-last_min);
series_add(serien[3],pitch[i]-pitch[last_min]);
series_add(serien[4],(pitch[i]-pitch[last_min])/(i-last_min));
last_max=i;
}
else
if (pitch[i-1]>pitch[i] && pitch[i]<pitch[i+1]) {
series_add(serien[5],pitch[i]);
series_add(serien[6],i-last_max);
series_add(serien[7],pitch[i]-pitch[last_max]);
series_add(serien[8],(pitch[i]-pitch[last_max])/(i-last_max));
last_min=i;
}
}
overall_max_ind = 1.0 * (overall_max_ind+1) / nframes;
if (nframes >1 && pitch[nframes-1]!=0)
series_add(serien[0],pitch[nframes-1]);
for (i=0; i<series_size; i++) {
getStatistics(features+fcount,serien[i]->series,serien[i]->nSeries); //pitch statistics features
fcount+=STATS;
}
// normierter Mittelwert, normierter Median, normiertes 1. Quartil, normiertes 3. Quartil
if (features[1]-features[2]) {
features[fcount++]=(features[0]-features[2])/(features[1]-features[2]);
features[fcount++]=(features[5]-features[2])/(features[1]-features[2]);
features[fcount++]=(features[6]-features[2])/(features[1]-features[2]);
features[fcount++]=(features[7]-features[2])/(features[1]-features[2]);
}
else {
features[fcount++]=0;
features[fcount++]=0;
features[fcount++]=0;
features[fcount++]=0;
}
features[fcount++]=1.0 * serien[1]->nSeries/nframes; // Anzahl Maxima pro Frames pro Segment
features[fcount++]=1.0 * serien[5]->nSeries/nframes; // Anzahl Minima pro Frames pro Segment
features[fcount++]=overall_max_ind; // Position des Maximums
n_voiced_frames=serien[0]->nSeries;
for (i=0;i<series_size;i++)
series_destroy(serien[i]);
}
getEnergy_and_MFCC(fex->mfcc, signal, fex->frame_len, &_energy_series, &_mfcc_series, &ser);
// ser=0;
if (ser) {
const int series_size = 8;
// @begin_change_johannes
//fextract_series_t *serien[series_size];
fextract_series_t *serien[8];
// @end_change_johannes
for (i=0;i<series_size;i++)
serien[i]=series_create(MAX_SERIES);
fcount+=getMFCCFeatures(features+fcount, _mfcc_series,ser);
/* Energie-Merkmale: */
last_max=last_min=0;
for (i=1;i<ser-1;i++) {
if (_energy_series[0][i-1]<_energy_series[0][i] && _energy_series[0][i]>_energy_series[0][i+1]) {
series_add(serien[0], _energy_series[0][i]); //Maxima
max_sum += _energy_series[0][i];
series_add(serien[1], i-last_min); // Laenge der Maxima
series_add(serien[2], _energy_series[0][i]-_energy_series[0][last_min]); // Groesse der Maxima
series_add(serien[3], (_energy_series[0][i]-_energy_series[0][last_min])/(i-last_min)); // Steigung der Maxima
last_max=i;
}
else
if (_energy_series[0][i-1]>_energy_series[0][i] && _energy_series[0][i]<_energy_series[0][i+1]) {
series_add(serien[4], _energy_series[0][i]); //Maxima
series_add(serien[5], i-last_max); // Laenge der Maxima
series_add(serien[6], _energy_series[0][i]-_energy_series[0][last_max]); // Groesse der Maxima
series_add(serien[7], (_energy_series[0][i]-_energy_series[0][last_min])/(i-last_max)); // Steigung der Maxima
last_max=i;
}
}
getStatistics(features+fcount,_energy_series[0],ser); // energy statistics
en_mean=features[fcount+1];
fcount+=STATS;
for (i=0;i<series_size;i++) {
getStatistics(features+fcount,serien[i]->series,serien[i]->nSeries);
fcount+=STATS;
}
max=serien[0]->nSeries;
for (j=1;j<3;j++) {
series_reset(serien[0]); series_reset(serien[1]);
for (i=1;i<ser-1;i++) {
if (_energy_series[j][i-1]<_energy_series[j][i] && _energy_series[j][i]>_energy_series[j][i+1]) {
series_add(serien[0], _energy_series[j][i]);
}
else
if (_energy_series[j][i-1]>_energy_series[j][i] && _energy_series[j][i]<_energy_series[j][i+1]) {
series_add(serien[1], _energy_series[j][i]);
}
}
getStatistics(features+fcount,_energy_series[j],ser); // energy delta statistics
fcount+=STATS;
getStatistics(features+fcount,serien[0]->series,serien[0]->nSeries);
fcount+=STATS;
getStatistics(features+fcount,serien[1]->series,serien[1]->nSeries);
fcount+=STATS;
}
features[fcount++]= max? (1.0 * max_sum/max) - en_mean:0; // Mittelwert der Maxima minus globaler Mittelwert
features[fcount++]= (1.0 * max)/ser; // Anzahl Maxima pro Frames pro Segment
for (i=0;i<series_size;i++)
series_destroy(serien[i]);
}
else {
for (;fcount < 141*STATS+9;fcount++)
features[fcount]=-1;
}
/* Voiced/Unvoiced-Merkmal: */
features[fcount++]=nframes ? (1.0 * n_voiced_frames)/nframes : -1; // Anzahl stimmloser Frames pro Frames pro Segment
/* Spectral Center of gravity-Merkmale: */
serie = getOnlyCoG(signal,fex->frame_len);
if (serie) {
getStatistics(features+fcount,serie->series,serie->nSeries);
series_destroy(serie);
fcount+=STATS;
}
else
for (;fcount< 142*STATS+10;fcount++)
features[fcount]=-1;
/* Dauer-Merkmal: */
features[fcount++]=fex->frame_len;
/* Harmonics-to-Noise-Ratio-Merkmale: */
hnr=hnr_calc(fex->hnr,signal,fex->frame_len);
nframes=fex->hnr->nframes;
if (!hnr)
for (;fcount<145*STATS+11;fcount++)
features[fcount]=-1.0;
else {
const int series_size=3;
// @begin_change_johannes
//fextract_series_t *serien[series_size];
fextract_series_t *serien[3];
// @end_change_johannes
pitch_frame_candidates_destroy(fex->hnr);
for (i=0;i<series_size;i++)
serien[i]=series_create(MAX_SERIES);
if (hnr[0]!=0)
series_add(serien[0],hnr[0]);
for (i=1;i<nframes-1;i++) {
if (hnr[i]!=-200) {
series_add(serien[0],hnr[i]);
if ((hnr[i-1]!=-200) && (hnr[i+1]!=200)) { // stimmlose Bereiche nicht beruecksichtigen
if (hnr[i-1]<hnr[i] && hnr[i]>hnr[i+1])
series_add(serien[1],hnr[i]);
else
if (hnr[i-1]>hnr[i] && hnr[i]<hnr[i+1])
series_add(serien[2],hnr[i]);
}
}
}
if (nframes >1 && hnr[nframes-1]!=0)
series_add(serien[0],hnr[nframes-1]);
for (i=0; i<series_size; i++) {
getStatistics(features+fcount,serien[i]->series,serien[i]->nSeries); //hnr statistics features
fcount+=STATS;
}
}
/* Aufraeumen... */
for (i=0;i<MFCC_FEATURES;i++)
if (_mfcc_series[i])
rs_free(_mfcc_series[i]);
if (_mfcc_series)
rs_free(_mfcc_series);
for (i=0;i<ENERGY_FEATURES;i++)
if (_energy_series[i])
rs_free(_energy_series[i]);
if (_energy_series)
rs_free(_energy_series);
if (pitch)
rs_free(pitch);
if (hnr)
rs_free(hnr);
/* for (;fcount<V1_N_FEATURES;fcount++) */
/* features[fcount]=0.0; */
if (fcount != V1_N_FEATURES)
rs_error("number of features is not correct (%d expected, %d calculated)!",V1_N_FEATURES,fcount);
return V1_N_FEATURES;
}
