int main() {
/* Initialization */
calc_shelving_coef();
calc_hamming_coef();
calc_bank_gain();
twidfftrad2_fr32(twiddle_table, WINDOW_LENGTH);
calc_dct_coef();
/* /Initialization */
int index;
for (index = 0; index < TOTAL_LENGTH; index++) {
input_fr[index] = float_to_fr32(test_input[index]);
}
pre_emphasis(input_fr, TOTAL_LENGTH);
// printf("pre emphasis done\n");
int frame_offset = 0;
fract32 frame_data[WINDOW_LENGTH];
int obs_length = 0;
int frame_num;
for (frame_num = 0; frame_num < FRAME_NUM; frame_num++) {
for (index = 0; index < WINDOW_LENGTH; index++) {
frame_data[index] = input_fr[index+frame_offset];
}
hamming(frame_data);
float energy = calc_energy(frame_data, WINDOW_LENGTH);
if (energy > 0.1) {
float mfcc[FEAT_NUM] = {0.0};
calc_mfcc(frame_data, mfcc_matrix[obs_length]);
obs_length++;
}
frame_offset += WINDOW_LENGTH/2;
}
int feat_num;
for (frame_num = 0; frame_num < obs_length; frame_num++) {
for (feat_num = 0; feat_num < FEAT_NUM; feat_num++) {
printf("%f\t", mfcc_matrix[frame_num][feat_num]);
}
printf("\n");
}
return 0;
}
int main() {
fract16 data[WINDOW_LENGTH];
int index;
for (index = 0; index < WINDOW_LENGTH; index++) {
data[index] = float_to_fr16(test_input[index]);
}
struct Coef shelving_coef = get_shelving_coef();
pre_emphasis(data, WINDOW_LENGTH, shelving_coef);
for (index = 0; index < WINDOW_LENGTH; ++index) {
float test = fr16_to_float(data[index]);
// printf("%f\n", fr16_to_float(data[index]));
}
return 0;
}
void Ambix_warpAudioProcessor::calcParams()
{
if (!_initialized)
{
sph_h_in.Init(in_order);
const String t_design_txt (t_design::des_3_240_21_txt);
// std::cout << t_design_txt << std::endl;
String::CharPointerType lineChar = t_design_txt.getCharPointer();
int n = 0; // how many characters been read
int numsamples = 0;
int i = 0;
int curr_n = 0;
int max_n = lineChar.length();
while (curr_n < max_n) { // check how many coordinates we have
double value;
sscanf(lineChar, "%lf\n%n", &value, &n);
lineChar += n;
curr_n += n;
numsamples++;
} // end parse numbers
numsamples = numsamples/3; // xyz
Carth_coord.resize(numsamples,3); // positions in cartesian coordinates
curr_n = 0;
lineChar = t_design_txt.getCharPointer();
// parse line for numbers again and copy to carth coordinate matrix
while (i < numsamples) {
double x,y,z;
sscanf(lineChar, "%lf%lf%lf%n", &x, &y, &z, &n);
Carth_coord(i,0) = x;
Carth_coord(i,1) = y;
Carth_coord(i,2) = z;
lineChar += n;
curr_n += n;
i++;
} // end parse numbers
// std::cout << "Coordinate size: " << Carth_coord.rows() << " x " << Carth_coord.cols() << std::endl;
// std::cout << Carth_coord << std::endl;
Sph_coord.resize(numsamples,2); // positions in spherical coordinates
Eigen::MatrixXd Sh_matrix;
Sh_matrix.setZero(numsamples,AMBI_CHANNELS);
int in_ambi_channels = (in_order+1)*(in_order+1);
for (int i=0; i < numsamples; i++)
{
Eigen::VectorXd Ymn(in_ambi_channels); // Ymn result
Sph_coord(i,0) = atan2(Carth_coord(i,1),Carth_coord(i,0)); // azimuth
Sph_coord(i,1) = atan2(Carth_coord(i,2),sqrt(Carth_coord(i,0)*Carth_coord(i,0) + Carth_coord(i,1)*Carth_coord(i,1))); // elevation
sph_h_in.Calc(Sph_coord(i,0),Sph_coord(i,1)); // phi theta
sph_h_in.Get(Ymn);
// std::cout << "Ymn Size: " << Ymn.size() << std::endl;
// std::cout << "Sh_matrix Size: " << Sh_matrix.size() << std::endl;
//Sh_matrix.row(i) = Ymn;
Sh_matrix.block(i,0,1,in_ambi_channels) = Ymn.transpose();
// std::cout << "Size: " << Sh_matrix.block(i,0,1,in_ambi_channels).size() << ": " << Sh_matrix.block(i,0,1,in_ambi_channels) << std::endl;
}
// Sh_matrix_inv.setZero();
//Sh_matrix_inv = (Sh_matrix.transpose()*Sh_matrix).inverse()*Sh_matrix.transpose(); // not working for dynamic input order
// if input order is different a better solving has to be used for the inverse:
Sh_matrix_inv = (Sh_matrix.transpose()*Sh_matrix).colPivHouseholderQr().inverse()*Sh_matrix.transpose();
// std::cout << "Size Sh_matrix: " << Sh_matrix.rows() << " x " << Sh_matrix.cols() << std::endl;
// std::cout << "Size Inverse: " << Sh_matrix_inv.rows() << " x " << Sh_matrix_inv.cols() << std::endl;
// std::cout << Sh_matrix_inv << std::endl;
_initialized = true;
}
if ( (phi_param != _phi_param) ||
(floor(phi_curve_param+0.5f) != floor(_phi_curve_param+0.5f)) ||
(theta_param != _theta_param) ||
(floor(theta_curve_param+0.5f) != floor(_theta_curve_param+0.5f)) ||
(in_order != _in_order) ||
(out_order != _out_order) ||
(floor(preemp_param+0.5f) != floor(_preemp_param+0.5f))
)
{
if (_out_order != out_order)
{
// init or reinit spherical harmonics
sph_h_out.Init(out_order);
}
Eigen::MatrixXd Sh_matrix_mod;
Sh_matrix_mod.setZero(Sph_coord.rows(),AMBI_CHANNELS);
Eigen::VectorXd pre_emphasis;
pre_emphasis.setOnes(Sph_coord.rows());
Eigen::MatrixXd Sph_coord_mod = Sph_coord;
// warping parameters - use inverse parameters! (see paper)
double phi_alpha = -((double)phi_param*1.8f - 0.9f);
double theta_alpha = -((double)theta_param*1.8f - 0.9f);
int out_ambi_channels = (out_order+1)*(out_order+1);
for (int i=0; i < Sph_coord_mod.rows(); i++)
{
if (theta_alpha != 0.)
{
// warp elevation
// as we are using elevation from -pi/2....pi/2
double mu = sin(Sph_coord(i,1));
if (theta_curve_param <= 0.5f)
{
// warp towards a pole
Sph_coord_mod(i,1) = asin((mu+theta_alpha)/(1+theta_alpha*mu));
// pre emphasis
pre_emphasis(i) = (1+theta_alpha*mu)/sqrt(1-theta_alpha*theta_alpha);
} else {
//warp towards equator (< 0.)or away from equator (> 0.)
// in the paper this parameter is called beta
// pre emphasis
pre_emphasis(i) = sqrt((1-abs(theta_alpha))*(1+abs(theta_alpha)*mu*mu))/(1-abs(theta_alpha)*mu*mu);
if (theta_alpha > 0.) {
Sph_coord_mod(i,1) = asin((theta_alpha-1+sqrt(pow(theta_alpha-1,2)+4*theta_alpha*pow(mu,2)))/(2*theta_alpha*mu));
} else {
Sph_coord_mod(i,1) = asin(((1+theta_alpha)*mu)/(1+theta_alpha*pow(mu,2)));
pre_emphasis(i) = 1/pre_emphasis(i); // this is ^sgn(beta)
}
}
}
/*
// this is not fully working yet
if (phi_alpha != 0.)
{
// warp azimuth
// as we are using azimuth from -pi....pi
double phi_temp = Sph_coord(i,0) * 0.5f;
double mu = sin(phi_temp);
if (phi_curve_param <= 0.5f)
{
// warp towards a pole
Sph_coord_mod(i,0) = asin((mu+phi_alpha)/(1+phi_alpha*mu));
// pre emphasis
pre_emphasis(i) *= (1+phi_alpha*mu)/sqrt(1-phi_alpha*phi_alpha);
} else {
//warp towards equator (< 0.)or away from equator (> 0.)
// in the paper this parameter is called beta
// pre emphasis
double pre_emph_temp = sqrt((1-abs(phi_alpha))*(1+abs(phi_alpha)*mu*mu))/(1-abs(phi_alpha)*mu*mu);
if (phi_alpha > 0.) {
Sph_coord_mod(i,0) = asin((phi_alpha-1+sqrt(pow(phi_alpha-1,2)+4*phi_alpha*pow(mu,2)))/(2*phi_alpha*mu));
pre_emphasis(i) *= pre_emph_temp;
} else {
Sph_coord_mod(i,0) = asin(((1+phi_alpha)*mu)/(1+phi_alpha*pow(mu,2)));
pre_emphasis(i) *= 1/pre_emph_temp; // this is ^sgn(beta)
}
}
Sph_coord_mod(i,0) *= 2;
}
*/
if (phi_alpha != 0.f)
{
// warp azimuth
// as we are using azimuth from -pi....pi
double phi_temp = Sph_coord(i,0) * 0.5f;
if (phi_curve_param <= 0.5f)
{
// warp towards poles
Sph_coord_mod(i,0) = asin((sin(phi_temp)+phi_alpha)/(1.f+phi_alpha*sin(phi_temp)));
} else {
//warp towards equator
if (phi_temp > 0)
{
Sph_coord_mod(i,0) = acos((cos(phi_temp*2)+phi_alpha)/(1+phi_alpha*cos(phi_temp*2))) / 2;
} else {
Sph_coord_mod(i,0) = -acos((cos(M_PI-(phi_temp-M_PI/2)*2)+phi_alpha)/(1+phi_alpha*cos(M_PI-(phi_temp-M_PI/2)*2))) / 2;
}
}
Sph_coord_mod(i,0) *= 2;
}
// rotate
// Sph_mod(i,0) = Sph(i,0);
Eigen::VectorXd Ymn(out_ambi_channels); // Ymn result
sph_h_out.Calc(Sph_coord_mod(i,0),Sph_coord_mod(i,1)); // phi theta
sph_h_out.Get(Ymn);
// Sh_matrix_mod.row(i) = Ymn;
Sh_matrix_mod.block(i,0,1,out_ambi_channels) = Ymn.transpose();
}
// apply preemphasis if wanted
if (preemp_param > 0.5f)
{
Sh_matrix_mod = Sh_matrix_mod * pre_emphasis.asDiagonal();
}
// calculate new transformation matrix
Sh_transf = Sh_matrix_inv * Sh_matrix_mod;
// threshold coefficients
for (int i = 0; i < Sh_transf.size(); i++)
{
if (abs(Sh_transf(i)) < 0.00001f)
Sh_transf(i) = 0.f;
}
_phi_param = phi_param;
_phi_curve_param = phi_curve_param;
_theta_param = theta_param;
_theta_curve_param = theta_curve_param;
_in_order = in_order;
_out_order = out_order;
_preemp_param = preemp_param;
}
}
void bob::ap::Ceps::operator()(const blitz::Array<double,1>& input,
blitz::Array<double,2>& ceps_matrix)
{
// Get expected dimensionality of output array
blitz::TinyVector<int,2> feature_shape = bob::ap::Ceps::getShape(input);
// Check dimensionality of output array
bob::core::array::assertSameShape(ceps_matrix, feature_shape);
int n_frames=feature_shape(0);
int shift_frame=0;
double last_frame_elem=0;
// Create the holder for the previous frame and make sure it's the same as the current frame
// Used by SSFC features computation
blitz::Array<double,1> _prev_frame_d;
_prev_frame_d.resize(m_cache_frame_d.shape());
// Create the temporary holder for SSFC features computation
blitz::Array<double,1> _temp_frame_d;
_temp_frame_d.resize(m_cache_frame_d.shape());
if (m_ssfc_features) {
//we are going to always process the next frame within the loop
shift_frame = 1;
// Init the first frame to the input
extractNormalizeFrame(input, 0, _prev_frame_d);
// Apply pre-emphasis
pre_emphasis(_prev_frame_d, last_frame_elem);
// Apply the Hamming window
hammingWindow(_prev_frame_d);
// Take the power spectrum of the first part of the FFT
powerSpectrumFFT(_prev_frame_d);
}
blitz::Range r1(0,m_n_ceps-1);
for (int i=0; i<n_frames; ++i)
{
// Init the current frame from the input, we process (i+1)th frame for SSFC features
extractNormalizeFrame(input, i+shift_frame, m_cache_frame_d);
// Update output with energy if required
if (m_with_energy)
ceps_matrix(i,(int)m_n_ceps) = logEnergy(m_cache_frame_d);
// Apply pre-emphasis
pre_emphasis(m_cache_frame_d, last_frame_elem);
// Apply the Hamming window
hammingWindow(m_cache_frame_d);
// Take the power spectrum of the first part of the FFT
// Note that after this call, we only operate on the first half of m_cache_frame_d array. The second half is ignored.
// powerSpectrumFFT changes first half+1 elements of m_cache_frame_d array
powerSpectrumFFT(m_cache_frame_d);
if (m_ssfc_features)
{
// retrieve the previous frame into our temp
_temp_frame_d = _prev_frame_d;
// remember the current frame for the next round, before we change current frame
_prev_frame_d = m_cache_frame_d;
// Computation of SSFC features:
// We take the previous frame and find the difference between values of current and previous frames
m_cache_frame_d -= _temp_frame_d;
// We compute norm2 for the difference as per SSFC features
m_cache_frame_d = blitz::pow2(m_cache_frame_d);
// Then, we can apply the filter and DCT later on
}
// Filter with triangular or rectangular filter bank (either in linear or Mel domain)
filterBank(m_cache_frame_d);
// Apply DCT kernel and update the output
blitz::Array<double,1> ceps_matrix_row(ceps_matrix(i,r1));
if (m_scfc_features)
// do not apply DCT on SCFC features
ceps_matrix_row = m_cache_filters(r1);
else
applyDct(ceps_matrix_row);
}
//compute the center of the cut-off frequencies
const int n_coefs = (m_with_energy ? m_n_ceps + 1 : m_n_ceps);
blitz::Range rall = blitz::Range::all();
blitz::Range ro0(0,n_coefs-1);
blitz::Range ro1(n_coefs,2*n_coefs-1);
blitz::Range ro2(2*n_coefs,3*n_coefs-1);
if (m_with_delta)
{
blitz::Array<double,2> ceps_matrix_0(ceps_matrix(rall,ro0));
blitz::Array<double,2> ceps_matrix_1(ceps_matrix(rall,ro1));
addDerivative(ceps_matrix_0, ceps_matrix_1);
if (m_with_delta_delta)
{
blitz::Array<double,2> ceps_matrix_2(ceps_matrix(rall,ro2));
addDerivative(ceps_matrix_1, ceps_matrix_2);
}
}
}
