void _fini()
#endif
{
#ifdef XTRACT_FFT
xtract_free_fft();
fftwf_cleanup();
#endif
}
void _fini()
#endif
{
xtract_free_fft();
}
map<char, vector<double> > soundFeatures::calcFeatures(string featString, int windowStart, int windowLength){
double mean, dev, zcr, hzcrr, kurtosis, skewness, variance, RMS, sharp, lpcc_order, loud;
double s_var ,s_mean, s_dev, s_kur, s_skew, s_centroid;
double spectrum[windowLength], autoCorr[windowLength+1], lpc[20], lpcc[15], mfcc[40], barkC[25];
int bark[26];
xtract_mel_filter mfccBank;
map <string, bool> calcFeat;
for(int k = 0; k < featString.size(); k++){
switch (featString[k]){
case 'a' :
case 'b' :
if(!calcFeat["zcr"]){
xtract[XTRACT_ZCR](wavData + windowStart, windowLength, NULL, &zcr);
calcFeat["zcr"] = true;
}
if(featString[k]== 'a') result['a'] = vector<double>{zcr};
if(featString[k]== 'b'){
double tempZRC;
double tempRes=0;
double tempN= windowLength / subWindow;
double tempAvg= zcr;
for (int i=0; i < windowLength ; i+=subWindow){
xtract[XTRACT_ZCR](wavData+i+windowStart, subWindow, NULL, &tempZRC);
if (tempZRC > 1.5 * tempAvg) tempRes++;
}
hzcrr = tempRes / (2*tempN);
result['b'] = vector<double>{hzcrr};
}
break;
case 'c':
case 'd':
if(!calcFeat["scalars"]){
xtract[XTRACT_MEAN](wavData + windowStart, windowLength, NULL, &mean);
xtract[XTRACT_VARIANCE](wavData + windowStart, windowLength, &mean, &variance);
xtract[XTRACT_STANDARD_DEVIATION](wavData + windowStart, windowLength, &variance, &dev);
calcFeat["scalars"] = true;
}
if(featString[k]== 'c'){
double temp_sca[2]={mean, dev};
xtract[XTRACT_KURTOSIS](wavData + windowStart, windowLength, temp_sca , &kurtosis);
result['c'] = vector<double>{kurtosis};
}
if(featString[k]== 'd'){
double temp_sca[2]={mean, dev};
xtract[XTRACT_SKEWNESS](wavData + windowStart, windowLength, temp_sca, &skewness);
result['d'] = vector<double>{skewness};
}
break;
case 'e':
case 'f':
case 'g':
case 'h':
case 'i':
case 'j':
case 'k':
if (!calcFeat["spectrum"]){
double temp1[4] = {(double)samplerate/windowLength, XTRACT_MAGNITUDE_SPECTRUM, 0.0f, 0.0f};
xtract_init_fft(windowLength, XTRACT_SPECTRUM);
xtract[XTRACT_SPECTRUM](wavData + windowStart, windowLength, &temp1[0], &spectrum[0]);
xtract_free_fft();
xtract[XTRACT_SPECTRAL_MEAN](spectrum, windowLength, NULL, &s_mean);
xtract[XTRACT_SPECTRAL_VARIANCE](spectrum, windowLength, &s_mean, &s_var);
xtract[XTRACT_SPECTRAL_STANDARD_DEVIATION](spectrum, windowLength, &s_var, &s_dev);
calcFeat["spectrum"] = true;
}
if(featString[k]== 'e') result['e'] = vector<double>{s_mean};
if(featString[k]== 'f') result['f'] = vector<double>{s_var};
if(featString[k]== 'g') result['g'] = vector<double>{s_dev};
{//radi odvajanja deklaracije temp unutar samo ovog bloka
double temp[2] = {s_mean, s_dev};
if(featString[k]== 'h'){
xtract[XTRACT_SPECTRAL_CENTROID](spectrum, windowLength, NULL, &s_centroid);
result['h'] = vector<double>{s_centroid};
}
if(featString[k]== 'i'){
xtract[XTRACT_SPECTRAL_KURTOSIS](spectrum, windowLength, temp, &s_kur);
result['i'] = vector<double>{s_kur};
}
if(featString[k]== 'j'){
xtract[XTRACT_SPECTRAL_SKEWNESS](spectrum, windowLength, temp, &s_skew);;
result['j'] = vector<double>{s_skew};
}
if(featString[k]== 'k'){
xtract[XTRACT_SHARPNESS](spectrum, windowLength/2, NULL, &sharp);
result['k'] = vector<double>{sharp};
}
}
break;
case 'l':
if (!calcFeat["spectrum"]){
double temp1[4] = {(double)samplerate/windowLength, XTRACT_MAGNITUDE_SPECTRUM, 0.0f, 0.0f};
xtract_init_fft(windowLength, XTRACT_SPECTRUM);
xtract[XTRACT_SPECTRUM](wavData + windowStart, windowLength, &temp1[0], &spectrum[0]);
xtract_free_fft();
calcFeat["spectrum"] = true;
}
xtract_init_bark(windowLength/2, samplerate, bark);
xtract[XTRACT_BARK_COEFFICIENTS](spectrum, windowLength/2, bark, barkC);
calcFeat["bark"] = true;
xtract[XTRACT_LOUDNESS](barkC, 10, NULL, &loud);
result['l'] = vector<double>{loud};
break;
case 'm':
xtract[XTRACT_RMS_AMPLITUDE](wavData + windowStart, windowLength, NULL, &RMS);
result['m'] = vector<double>{RMS};
break;
case 'n':
xtract[XTRACT_AUTOCORRELATION](wavData + windowStart, windowLength, NULL, autoCorr);
xtract[XTRACT_LPC](autoCorr, 20, NULL, lpc);
//lpcc_order = 15;
//xtract[XTRACT_LPCC](lpc + 10, 10, &lpcc_order, lpcc);
//result['n'] = vector<double>(lpcc, lpcc + (int)lpcc_order);
result['n'] = vector<double>(lpc+20, lpc + 40);
break;
case 'o':
if (!calcFeat["spectrum"]){
double temp1[4] = {(double)samplerate/windowLength, XTRACT_MAGNITUDE_SPECTRUM, 0.0f, 0.0f};
xtract_init_fft(windowLength, XTRACT_SPECTRUM);
xtract[XTRACT_SPECTRUM](wavData + windowStart, windowLength, &temp1[0], &spectrum[0]);
xtract_free_fft();
calcFeat["spectrum"] = true;
}
mfccBank.n_filters = 40;
mfccBank.filters = (double **)malloc(40 * sizeof(double *));
for(int i = 0; i < 40; i++)
mfccBank.filters[i] = (double *)malloc(windowLength * sizeof(double));
xtract_init_mfcc(windowLength/2, samplerate/2, XTRACT_EQUAL_GAIN, 20, samplerate/2, mfccBank.n_filters, mfccBank.filters);
xtract[XTRACT_MFCC](spectrum, windowLength/2, &mfccBank, mfcc);
for(int i = 0; i < 40; ++i)
free(mfccBank.filters[i]);
free(mfccBank.filters);
result['o'] = vector<double>(mfcc, mfcc + 40);
break;
case 'p':
if (!calcFeat["bark"]){
if (!calcFeat["spectrum"]){
double temp1[4] = {(double)samplerate/windowLength, XTRACT_MAGNITUDE_SPECTRUM, 0.0f, 0.0f};
xtract_init_fft(windowLength, XTRACT_SPECTRUM);
xtract[XTRACT_SPECTRUM](wavData + windowStart, windowLength, &temp1[0], &spectrum[0]);
xtract_free_fft();
calcFeat["spectrum"] = true;
}
xtract_init_bark(windowLength/2, samplerate, bark);
xtract[XTRACT_BARK_COEFFICIENTS](spectrum, windowLength/2, bark, barkC);
calcFeat["bark"] = true;
}
result['p'] = vector<double>(barkC, barkC + 24);
break;
}
}
return result;
}
double* extractall(double* wavetable, long numFrames, double samplerate){
long size = 38+5*BLOCKSIZE+MFCC_FREQ_BANDS;
double *FVec = (double*) malloc(size*sizeof(double));
long count=0;
long i,n,j,k;
int rv = XTRACT_SUCCESS;
xtract_mel_filter mel_filters;
/* get the F0 */
xtract[XTRACT_WAVELET_F0](wavetable, BLOCKSIZE, &samplerate, &f0);
FVec[count++] = f0;
//* get the F0 as a MIDI note
xtract[XTRACT_MIDICENT](NULL, 0, &f0, &midicents);
FVec[count++] = midicents;
//* get the mean of the input
xtract[XTRACT_MEAN](wavetable, BLOCKSIZE, NULL, &mean);
FVec[count++] = mean;
//MY-VARIANCE
xtract[XTRACT_VARIANCE](wavetable, BLOCKSIZE, &mean, &variance);
FVec[count++] = variance;
//MY-STANDARD DEVIATION
xtract[XTRACT_STANDARD_DEVIATION](wavetable, BLOCKSIZE, &variance, &stdDeviation);
FVec[count++] = stdDeviation;
//MY-AVERAGE DEVIATION
xtract[XTRACT_AVERAGE_DEVIATION](wavetable, BLOCKSIZE, &mean, &avgDeviation);
FVec[count++] = avgDeviation;
//MY-SKEWNESS
argd[0] = mean;
argd[1] = stdDeviation;
xtract[XTRACT_SKEWNESS](wavetable, BLOCKSIZE, argd, &skewness);
FVec[count++] = skewness;
//MY-KURTOSIS
xtract[XTRACT_KURTOSIS](wavetable, BLOCKSIZE, argd, &kurtosis);
FVec[count++] = kurtosis;
//MY-ZERO CROSSING RATE
xtract[XTRACT_ZCR](wavetable, BLOCKSIZE, NULL, &zcr);
FVec[count++] = zcr;
//MY-RMS AMPLITUDE
xtract[XTRACT_RMS_AMPLITUDE](wavetable, BLOCKSIZE, NULL, &rmsAmplitude);
FVec[count++] = rmsAmplitude;
//MY-HIGHEST VALUE
xtract[XTRACT_HIGHEST_VALUE](wavetable, BLOCKSIZE, NULL, &highestValue);
FVec[count++] = highestValue;
//MY-SUM
xtract[XTRACT_SUM](wavetable, BLOCKSIZE, NULL, &sum);
FVec[count++] = sum;
//MY-CREST
argd[0] = mean;
argd[1] = highestValue;
xtract[XTRACT_CREST](wavetable, BLOCKSIZE, argd, &crest);
FVec[count++] = crest;
//* get the lowest value in the input
argd[0] = -.5;
rv = xtract[XTRACT_LOWEST_VALUE](wavetable, BLOCKSIZE, argd, &lowest);
FVec[count++] = lowest;
//MY-AMDF
xtract[XTRACT_AMDF](wavetable, BLOCKSIZE, NULL, amdf);
for(k=0;k<BLOCKSIZE;k++)
FVec[i++] = amdf[k];
//MY-ASDF
xtract[XTRACT_ASDF](wavetable, BLOCKSIZE, NULL, asdf);
for(k=0;k<BLOCKSIZE;k++)
FVec[i++] = asdf[k];
//MY-DCT
xtract[XTRACT_DCT](wavetable, BLOCKSIZE, NULL, dct);
for(k=0;k<BLOCKSIZE;k++)
FVec[i++] = dct[k];
//* create the window function
window = xtract_init_window(BLOCKSIZE, XTRACT_HANN);
xtract_windowed(wavetable, BLOCKSIZE, window, windowed);
xtract_free_window(window);
//* get the spectrum
argd[0] = SAMPLERATE / (double)BLOCKSIZE;
argd[1] = XTRACT_MAGNITUDE_SPECTRUM;
argd[2] = 0.f; //* DC component - we expect this to zero for square wave
argd[3] = 0.f; //* No Normalisation
xtract_init_fft(BLOCKSIZE, XTRACT_SPECTRUM);
xtract[XTRACT_SPECTRUM](windowed, BLOCKSIZE, &argd[0], spectrum);
xtract_free_fft();
//MY_BARK
// xtract_init_bark(BLOCKSIZE,samplerate,barkCoeff);
// xtract[XTRACT_BARK_COEFFICIENTS](spectrum,BLOCKSIZE,barkCoeff,bark);
//MY-SPECTRAL MEAN
xtract[XTRACT_SPECTRAL_MEAN](spectrum, BLOCKSIZE, NULL, &specMean);
FVec[count++] = specMean;
//MY-SPECTRAL VARIANCE
xtract[XTRACT_SPECTRAL_VARIANCE](spectrum, BLOCKSIZE, &specMean, &specVariance);
FVec[count++] = specVariance;
//MY-SPECTRAL STANDARD DEVIATION
xtract[XTRACT_SPECTRAL_STANDARD_DEVIATION](spectrum, BLOCKSIZE, &specMean, &specStdDeviation);
FVec[count++] = specStdDeviation;
//xtract[XTRACT_SPECTRAL_AVERAGE_DEVIATION](spectrum, BLOCKSIZE, &specMean, &specAvgDeviation);
//FVec[count++] = specAvgDeviation;
//MY-SPECTRAL SKEWNESS
argd[0] = specMean;
argd[1] = specStdDeviation;
xtract[XTRACT_SPECTRAL_SKEWNESS](spectrum, BLOCKSIZE, argd, &specSkewness);
FVec[count++] = specSkewness;
//MY-SPECTRAL KURTOSIS
xtract[XTRACT_SPECTRAL_KURTOSIS](spectrum, BLOCKSIZE, argd, &specKurtosis);
FVec[count++] = specKurtosis;
xtract[XTRACT_SPECTRAL_CENTROID](spectrum, BLOCKSIZE, NULL, ¢roid);
FVec[count++] = centroid;
//MY-POWER
xtract[XTRACT_POWER](spectrum, BLOCKSIZE, NULL, &power);
FVec[count++] = power;
//MY-SHARPNESS
xtract[XTRACT_SHARPNESS](spectrum, BLOCKSIZE, NULL, &sharpness);
FVec[count++] = sharpness;
//MY-SPECTRAL SLOPE
xtract[XTRACT_SPECTRAL_SLOPE](spectrum, BLOCKSIZE/2, NULL, &specSlope);
FVec[count++] = specSlope;
//MY-HPS
xtract[XTRACT_HPS](spectrum, BLOCKSIZE, NULL, &hps);
FVec[count++] = hps;
//MY-FLATNESS
xtract[XTRACT_FLATNESS](spectrum, BLOCKSIZE, NULL, &flatness);
FVec[count++] = flatness;
//MY-TONALITY
xtract[XTRACT_TONALITY](NULL, NULL, &flatness, &tonality);
FVec[count++] = tonality;
argd[0] = SAMPLERATE / (double)BLOCKSIZE;
argd[1] = 10.0; //* peak threshold as % of maximum peak
xtract[XTRACT_PEAK_SPECTRUM](spectrum, BLOCKSIZE / 2, argd, peaks);
argd[0] = f0;
argd[1] = .3; //* harmonic threshold
xtract[XTRACT_HARMONIC_SPECTRUM](peaks, BLOCKSIZE, argd, harmonics);
//MY-ODD EVEN RATIO
xtract[XTRACT_ODD_EVEN_RATIO](harmonics, BLOCKSIZE, NULL, &oddEvenRatio);
FVec[count++] = oddEvenRatio;
//MY-TRISTIMULUS-1
xtract[XTRACT_TRISTIMULUS_1](harmonics, BLOCKSIZE, NULL, &tristimulus1);
FVec[count++] = tristimulus1;
//MY-TRISTIMULUS-2
xtract[XTRACT_TRISTIMULUS_2](harmonics, BLOCKSIZE, NULL, &tristimulus2);
FVec[count++] = tristimulus2;
//MY-TRISTIMULUS-2
xtract[XTRACT_TRISTIMULUS_3](harmonics, BLOCKSIZE, NULL, &tristimulus3);
FVec[count++] = tristimulus3;
//MY-NOISINESS
xtract[XTRACT_NOISINESS](NULL, NULL, harmonics, &noisiness);
FVec[count++] = noisiness;
//MY-SPECTRAL INHARMONICITY
// xtract[XTRACT_SPECTRAL_INHARMONICITY](peaks, BLOCKSIZE, &f0, &specInharmonicity);
// FVec[count++] = specInharmonicity;
//MY-SPERAD
// xtract[XTRACT_SPREAD](spectrum, BLOCKSIZE, NULL, &spread);
// FVec[count++] = spread;
//MY-AUTOCORRELATION and AUTOCORRELATION FFT
// xtract[XTRACT_AUTOCORRELATION](wavetable, BLOCKSIZE, NULL, &autocorrelation);
// FVec[count++] = autocorrelation;
// printf("\nAutocorrelation : %f\n", autocorrelation);
//xtract[XTRACT_AUTOCORRELATION_FFT](wavetable, BLOCKSIZE, NULL, &autocorrelationFFT);
//FVec[count++] = autocorrelationFFT;
for(k=0;k<BLOCKSIZE;k++)
FVec[i++] = spectrum[k];
//* compute the MFCCs
mel_filters.n_filters = MFCC_FREQ_BANDS;
mel_filters.filters = (double **)malloc(MFCC_FREQ_BANDS * sizeof(double *));
for(n = 0; n < MFCC_FREQ_BANDS; ++n)
{
mel_filters.filters[n] = (double *)malloc(BLOCKSIZE * sizeof(double));
}
xtract_init_mfcc(BLOCKSIZE >> 1, SAMPLERATE >> 1, XTRACT_EQUAL_GAIN, MFCC_FREQ_MIN, MFCC_FREQ_MAX, mel_filters.n_filters, mel_filters.filters);
xtract_mfcc(spectrum, BLOCKSIZE >> 1, &mel_filters, mfccs);
for(k=0;k<MFCC_FREQ_BANDS;k++)
FVec[i++] = mfccs[k];
//* compute Spectral Flux
argd[0] = SAMPLERATE / (BLOCKSIZE/2);
argd[1] = XTRACT_MAGNITUDE_SPECTRUM;
argd[2] = 0.f; //* DC component
argd[3] = 0.f; //* No Normalisation
xtract_init_fft(BLOCKSIZE/2, XTRACT_SPECTRUM);
xtract_features_from_subframes(wavetable, BLOCKSIZE, XTRACT_SPECTRUM, argd, subframes);
xtract_difference_vector(subframes, BLOCKSIZE, NULL, difference);
argd[0] = 1.0; //* norm order
argd[1] = XTRACT_POSITIVE_SLOPE; //* positive slope
xtract_flux(difference, BLOCKSIZE/2, argd, &flux);
FVec[i++] = flux;
for(k=0;k<BLOCKSIZE;k++)
FVec[i++] = peaks[k];
/*
printf("\nF0: %f\n", f0);
printf("\nMIDI cents: %f\n", midicents);
printf("\nInput mean = %.5f\t%.5f\n", mean); //* We expect this to be zero for a square wave
printf("\nVariance = %f\n",variance);
printf("\nStandard Deviation = %f\n",stdDeviation);
printf("\nAverage Deviation = %f\n",avgDeviation);
printf("\nSkewness = %f\n",skewness);
printf("\nKurtosis = %f\n",kurtosis);
printf("\nZero Crossing Rate = %f\n",zcr);
printf("\nRMS Amplitude = %f\n",rmsAmplitude);
printf("\nHighest Value = %f\n",highestValue);
printf("\nSum = %f\n", sum);
printf("\nCrest = %f\n", crest);
printf("\nLowest value = %.6f\n\n", lowest);
for(n=0;n<BLOCKSIZE;n+=50)
printf("\nAMDF[%d]: %f",n,amdf[n]);
for(n=0;n<BLOCKSIZE;n+=50)
printf("\nASDF[%d]: %f",n,asdf[n]);
for(n=0;n<BLOCKSIZE;n+=50)
printf("\nDCT[%d]: %f",n,dct[n]);
// for(n=0;n<BLOCKSIZE;n+=1000)
// printf("\nBARK[%d]: %f",n,bark[n]);
printf("\nSpectral Mean: %f\n", specMean);
printf("\nSpectral Variance: %f\n", specVariance);
printf("\nSpectral Standard Deviation: %f\n", specStdDeviation);
//printf("\nSpectral Average Deviation: %f\n", specAvgDeviation);
printf("\nSpectral Skewness: %f\n", specSkewness);
printf("\nSpectral Kurtosis: %f\n", specKurtosis);
printf("\nSpectral Centroid: %f\n", centroid);
printf("\nPower = %f\n",power);
printf("\nSharpness = %f\n",sharpness);
printf("\nSpectral Slope = %f\n",specSlope);
printf("\nHPS = %f\n",hps);
printf("\nFlatness: %f\n", flatness);
printf("\nTonality: %f\n", tonality);
printf("\nOdd Even Ratio : %f\n", oddEvenRatio);
printf("\nTristimulus-1 : %f\n", tristimulus1);
printf("\nTristimulus-2 : %f\n", tristimulus2);
printf("\nTristimulus-3 : %f\n", tristimulus3);
printf("\nNoisiness : %f\n", noisiness);
// printf("\nSpectral Inharmonicity : %f\n", specInharmonicity);
// printf("\nSpread : %f\n", spread);
// printf("\nAutocorrelation fft : %f\n",autocorrelationFFT);
printf("\nSpectrum:\n");
for(n = 0; n < (BLOCKSIZE >> 1); n+=1000)
{
printf("freq: %.1f\tamp: %.6f", spectrum[n + (BLOCKSIZE >> 1)], spectrum[n]);
if (peaks[n + (BLOCKSIZE >> 1)] != 0.f)
{
printf("\tpeak:: freq: %.1f\tamp: %.6f\n", peaks[n + (BLOCKSIZE >> 1)], peaks[n]);
}
else
{
printf("\n");
}
}
printf("\n");
printf("MFCCs:\n");
for(n = 0; n < MFCC_FREQ_BANDS; ++n)
{
printf("band: %d\t", n);
if(n < 10) {
printf("\t");
}
printf("coeff: %f\n", mfccs[n]);
}
printf("Flux: %f\n", flux);*/
//}
// cleanup
for(n = 0; n < MFCC_FREQ_BANDS; ++n)
{
free(mel_filters.filters[n]);
}
free(mel_filters.filters);
return FVec;
}
