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;
}
static void *xtract_new(t_symbol *me, t_int argc, t_atom *argv) {
t_xtract_tilde *x = (t_xtract_tilde *)pd_new(xtract_class);
xtract_mel_filter *mf;
t_int n, N, M, f, F,
n_args,
type;
t_float *argv_max;
t_symbol *arg1;
xtract_function_descriptor_t *fd;
char *p_name,
*p_desc,
*author;
int year;
p_name = p_desc = author = NULL;
n_args = type = 0;
f = F = XTRACT_FEATURES;
N = BLOCKSIZE;
x->argv = NULL;
x->argv_type = 0;
x->done_init = 0;
x->is_scalar = 0;
x->is_subframe = 0;
x->feature = -1;
/* Allocate data area */
x->data = (double *)getbytes(N * sizeof(double));
x->result = (double *)getbytes(N * sizeof(double));
/* Parse arguments */
if(argc){
arg1 = atom_getsymbol(argv);
if(arg1 == gensym("subframe"))
x->is_subframe = 1;
else
x->feature_name = atom_getsymbol(argv);
}
else {
post("xtract~: No arguments given");
return (void *)x;
}
if(argc > 1){
if(x->is_subframe)
x->feature_name = atom_getsymbol(argv+1);
else
N = atom_getint(argv+1);
}
if(argc > 2)
N = atom_getint(argv+2);
x->init_blocksize = N;
M = N >> 1;
/* get function descriptors */
fd = (xtract_function_descriptor_t *)xtract_make_descriptors();
/* iterate over descriptors */
while(f--){
/* map creation arg to feature */
if(x->feature_name == gensym(fd[f].algo.name)){
x->feature = f;
break;
}
}
if(x->feature == -1)
post("xtract~: feature not found: %s", x->feature_name->s_name);
/* allocate memory for feature arguments */
n_args = fd[f].argc;
type = fd[f].argv.type;
x->argv_type = type;
if(n_args){
for(n = 0; n < n_args; n++){
argv_max = &fd[f].argv.max[n];
/*post("Argument %d, max: %.2f", n, *argv_max); */
}
if(type == XTRACT_MEL_FILTER){
x->memory.argv = (size_t)(n_args * sizeof(xtract_mel_filter));
x->argv = (xtract_mel_filter *)getbytes(x->memory.argv);
}
else if(type == XTRACT_INT){
x->memory.argv = (size_t)(n_args * sizeof(t_int));
x->argv = (t_int *)getbytes(x->memory.argv);
}
else if (type == XTRACT_FLOAT){
x->memory.argv = (size_t)(n_args * sizeof(t_float));
x->argv = (t_float *)getbytes(x->memory.argv);
}
else
x->memory.argv = 0;
}
p_name = fd[f].algo.p_name;
p_desc = fd[f].algo.p_desc;
author = fd[f].algo.author;
year = fd[f].algo.year;
if(argc){
if(strcmp(p_name, ""))
post("xtract~: %s", p_name );
if(strcmp(p_desc, ""))
post("xtract~: %s", p_desc );
if(strcmp(author, "") && year)
post("xtract~: %s(%d)", author, year);
}
/* Adjust frame size if we are using subframe features */
if(x->is_subframe)
N = M;
post("xtract~: window size: %d", N);
/* do init if needed */
if(x->feature == XTRACT_MFCC){
mf = x->argv;
mf->n_filters = 20;
post("xtract~: mfcc: filters = %d",
((xtract_mel_filter *)x->argv)->n_filters);
mf->filters =
(t_float **)getbytes(mf->n_filters * sizeof(t_float *));
for(n = 0; n < mf->n_filters; n++)
mf->filters[n] = (float *)getbytes(N * sizeof(float));
xtract_init_mfcc(N, NYQUIST, XTRACT_EQUAL_GAIN, 80.0f,
18000.0f, mf->n_filters, mf->filters);
x->done_init = 1;
}
else if(x->feature == XTRACT_BARK_COEFFICIENTS){
xtract_init_bark(N, NYQUIST, x->argv);
x->done_init = 1;
}
else if(x->feature == XTRACT_WINDOWED){
x->window = xtract_init_window(N, XTRACT_HANN);
x->argv = x->window;
x->done_init = 1;
}
/* Initialise fft_plan if required */
if(x->feature == XTRACT_AUTOCORRELATION_FFT ||
x->feature == XTRACT_SPECTRUM ||
x->feature == XTRACT_DCT){
xtract_init_fft(N, x->feature);
x->done_init = 1;
}
if(fd[f].is_scalar)
x->is_scalar = 1;
/*
if(x->feature == XTRACT_AUTOCORRELATION ||
x->feature == XTRACT_AUTOCORRELATION_FFT ||
x->feature == XTRACT_MFCC || x->feature == XTRACT_AMDF ||
x->feature == XTRACT_ASDF|| x->feature == XTRACT_DCT ||
x->feature == XTRACT_BARK_COEFFICIENTS ||
x->feature == XTRACT_SPECTRUM ||
x->feature == XTRACT_PEAK_SPECTRUM ||
x->feature == XTRACT_HARMONIC_SPECTRUM ||
x->feature == XTRACT_LPC ||
x->feature == XTRACT_LPCC ||
x->feature == XTRACT_WINDOWED)
x->feature_type = XTRACT_VECTOR;
*/
/* else if (x->feature == XTRACT_FLUX || x->feature == XTRACT_ATTACK_TIME ||
x->feature == XTRACT_DECAY_TIME || x->feature == XTRACT_DIFFERENCE_VECTOR)
x->feature_type = XTRACT_DELTA; */
/*
else x->feature_type = XTRACT_SCALAR;
*/
/* argv through right inlet */
inlet_new(&x->x_obj, &x->x_obj.ob_pd, gensym("list"), gensym("list"));
/* if feature is vector, create signal out */
if(!x->is_scalar)
outlet_new(&x->x_obj, &s_signal);
/* otherwise: float */
else
outlet_new(&x->x_obj, &s_float);
if(x->is_scalar && x->is_subframe)
post(
"xtract~: warning: subframes not yet supported for scalar features");
/* free the function descriptors */
xtract_free_descriptors(fd);
return (void *)x;
}
static void *xtract_tilde_new(t_symbol *me, t_int argc, t_atom *argv) {
t_symbol *tmp;
t_xtract_tilde *x = (t_xtract_tilde *)newobject(xtract_tilde_class);
xtract_mel_filter *mf;
t_int n, N, f, F, n_args, type;
t_float *argv_max;
xtract_function_descriptor_t *fd;
char *p_name, *p_desc, *author;
int year;
tmp = NULL;
p_name = p_desc = author = NULL;
n_args = type = x->feature = 0;
f = F = XTRACT_FEATURES;
N = BLOCKSIZE;
x->argv = NULL;
x->done_init = 0;
if(argc)
tmp = argv[0].a_w.w_sym; /*atom_getsymbol(argv); */
if(argc > 1)
N = (t_int)argv[1].a_w.w_long;
x->init_blocksize = N;
/* get function descriptors */
fd = (xtract_function_descriptor_t *)xtract_make_descriptors();
/* iterate over descriptors */
while(f--){
/* map creation arg to feature */
if(tmp == gensym(fd[f].algo.name)){
x->feature = f;
break;
}
}
/* allocate memory for feature arguments */
n_args = fd[f].argc;
type = fd[f].argv.type;
if(n_args){
for(n = 0; n < n_args; n++){
argv_max = &fd[f].argv.max[n];
//post("Argument %d, max: %.2f", n, *argv_max);
}
if(type == XTRACT_MEL_FILTER){
x->memory.argv = (size_t)(n_args * sizeof(xtract_mel_filter));
x->argv = (xtract_mel_filter *)getbytes(x->memory.argv);
}
else if(type == XTRACT_INT){
x->memory.argv = (size_t)(n_args * sizeof(t_int));
x->argv = (t_int *)getbytes(x->memory.argv);
}
else if (type == XTRACT_FLOAT){
x->memory.argv = (size_t)(n_args * sizeof(t_float));
x->argv = (t_float *)getbytes(x->memory.argv);
}
else
x->memory.argv = 0;
}
p_name = fd[f].algo.p_name;
p_desc = fd[f].algo.p_desc;
author = fd[f].algo.author;
year = fd[f].algo.year;
if(argc){
if(strcmp(p_name, ""))
post("xtract~: %s", p_name );
if(strcmp(p_desc, ""))
post("xtract~: %s", p_desc );
if(strcmp(author, "") && year)
post("xtract~: %s(%d)", author, year);
}
else
post("xtract~: No arguments given");
/* do init if needed */
if(x->feature == XTRACT_MFCC){
mf = x->argv;
mf->n_filters = 20;
post("xtract~: mfcc: filters = %d",
((xtract_mel_filter *)x->argv)->n_filters);
mf->filters =
(t_float **)getbytes(mf->n_filters * sizeof(t_float *));
for(n = 0; n < mf->n_filters; n++)
mf->filters[n] = (float *)getbytes(N * sizeof(float));
xtract_init_mfcc(N, NYQUIST, XTRACT_EQUAL_GAIN, 80.0f,
18000.0f, mf->n_filters, mf->filters);
}
else if(x->feature == XTRACT_BARK_COEFFICIENTS)
xtract_init_bark(N, NYQUIST, x->argv);
/* Initialise fft_plan if required */
if(x->feature == XTRACT_AUTOCORRELATION_FFT ||
x->feature == XTRACT_SPECTRUM ||
x->feature == XTRACT_DCT){
xtract_init_fft(N, x->feature);
x->done_init = 1;
}
if(x->feature == XTRACT_AUTOCORRELATION ||
x->feature == XTRACT_AUTOCORRELATION_FFT ||
x->feature == XTRACT_MFCC || x->feature == XTRACT_AMDF ||
x->feature == XTRACT_ASDF|| x->feature == XTRACT_DCT ||
x->feature == XTRACT_BARK_COEFFICIENTS ||
x->feature == XTRACT_SPECTRUM ||
x->feature == XTRACT_PEAK_SPECTRUM ||
x->feature == XTRACT_HARMONIC_SPECTRUM)
x->feature_type = XTRACT_VECTOR;
else if (x->feature == XTRACT_FLUX || x->feature == XTRACT_ATTACK_TIME ||
x->feature == XTRACT_DECAY_TIME || x->feature == XTRACT_DELTA)
x->feature_type = XTRACT_DELTA;
else x->feature_type = XTRACT_SCALAR;
/* argv through right inlet */
inlet_new((t_pxobject *)x, "list");
/* DSP inlet */
dsp_setup((t_pxobject *)x, 1);
/* if feature is vector, create signal out */
if(x->feature_type == XTRACT_VECTOR)
outlet_new((t_pxobject *)x, "signal");
/* otherwise: float */
else
x->outlet = floatout((t_pxobject *)x);
/* free the function descriptors */
xtract_free_descriptors(fd);
return (void *)x;
}
