/*
* NEURAL_NET neural network simulation function.
*
* Generated by Neural Network Toolbox function genFunction, 18-Mar-2016 18:06:53.
*
* [y1] = neural_net(x1) takes these arguments:
* x = 12xQ matrix, input #1
* and returns:
* y = 5xQ matrix, output #1
* where Q is the number of samples.
* Arguments : const float x1[1488]
* float b_y1[620]
* Return Type : void
*/
void neural_net(const float x1[1488], float b_y1[620])
{
float xp1[1488];
static const double dv0[12] = { 16.132376, -41.92286, -11.763175, 0.26042256,
-48.499973, -10.59572, -11.150261, -50.75682, -17.053158, -20.999031,
-31.260366, -16.086512 };
static const double dv1[12] = { 0.0872760507447396, 0.0583581858719483,
0.0496593431136582, 0.0621090054882946, 0.0398973259870547,
0.0429406458505015, 0.0418338504206415, 0.0292683416538594,
0.0406196264042357, 0.0324026540430679, 0.0488586616635397,
0.0444680174345756 };
double dv2[620];
double dv3[1240];
float fv0[1240];
int i0;
int i1;
float f0;
int i2;
static const float fv1[120] = { -0.482139856F, 0.268292874F, -0.398598433F,
-0.492702723F, -5.28534F, -1.3607614F, -1.31324863F, -2.90275884F,
0.57084167F, 0.643586755F, 0.54507637F, 0.229064658F, 0.454308599F,
-1.42380536F, -1.41992974F, 3.64180279F, 0.14793165F, 1.11103225F,
-0.886625707F, -1.1550293F, 0.21188508F, 2.01851439F, -0.85992837F,
-1.96682489F, -0.0088577373F, 3.95469904F, 0.961833417F, 2.36891127F,
-0.709682882F, -2.43880248F, 0.912824154F, 1.19552827F, -0.754117072F,
-0.0282568727F, -1.58362544F, -0.450968713F, -1.03895199F, 0.997594237F,
-0.53181994F, -0.204281747F, -0.679035425F, -0.652160525F, -1.03224313F,
0.574654937F, -0.419399559F, -1.09406209F, -1.95429432F, -1.64093769F,
0.616113186F, 1.53872609F, 0.877005696F, -0.772606075F, 0.965004742F,
-0.708744109F, -0.699137F, 0.178787217F, 0.661658704F, 1.50458348F,
-1.11914206F, 2.82313919F, 1.04887187F, 1.11462772F, 3.57028317F,
1.90066338F, 1.40841186F, 0.145877808F, -0.734749F, 1.91204357F,
-0.942125857F, 0.125358075F, 0.803959668F, -2.41720366F, -0.863904297F,
0.142200008F, -0.480167896F, -3.39202547F, -2.81302428F, -2.2363379F,
0.470471919F, -1.23218513F, -0.235131681F, -3.10341072F, -2.91667509F,
-0.866841495F, 0.148358867F, -0.961966038F, -0.191483766F, -2.43185115F,
0.39682892F, -1.77321279F, -0.733736455F, 2.46021F, 1.28283262F, 1.75353122F,
0.896416F, 2.57278728F, 1.11794055F, 3.0823822F, -0.0531957038F,
-1.96753991F, 0.432154804F, -0.0155567462F, -2.25758934F, 1.58146751F,
-0.0868694782F, -2.00317669F, 0.564771533F, -2.29165697F, 0.0372338742F,
1.49811876F, -0.943119347F, -1.67992508F, 0.932955503F, -2.28386927F,
2.41840959F, 0.486109674F, 2.88024926F, -0.583868921F, 1.52614343F,
1.9853704F };
float fv2[1240];
float fv3[620];
static const float fv4[50] = { 1.92884886F, 0.0791850165F, 0.606354833F,
0.8203246F, -0.852406919F, -2.10267973F, -0.585249543F, -1.96586227F,
5.016222F, -0.959461629F, -3.65689325F, 0.844967902F, 4.3082037F,
1.28906775F, -3.90302444F, 0.817251503F, -3.68290186F, -0.165898442F,
3.55818534F, -1.68369067F, -5.09043407F, 0.936949968F, 2.58750296F,
2.75450921F, 0.50600791F, 5.61685228F, 4.00498772F, -4.83895683F,
1.23465705F, -3.98150063F, -4.74256086F, 4.84532976F, -3.24344039F,
0.342350513F, 4.4398303F, -4.56724262F, 3.84799695F, -1.66734827F,
2.45293832F, -0.163840726F, -1.40159512F, 0.639144599F, 1.05117142F,
-0.097580716F, 1.9411335F, 3.98067117F, -2.86589026F, -3.9957943F,
-1.48145068F, 2.93353248F };
float fv5[620];
/* ===== NEURAL NETWORK CONSTANTS ===== */
/* Input 1 */
/* Layer 1 */
/* Layer 2 */
/* Output 1 */
/* ===== SIMULATION ======== */
/* Dimensions */
/* samples */
/* Input 1 */
mapminmax_apply(x1, dv1, dv0, xp1);
/* Layer 1 */
/* Layer 2 */
b_repmat(dv2);
repmat(dv3);
for (i0 = 0; i0 < 10; i0++) {
for (i1 = 0; i1 < 124; i1++) {
f0 = 0.0F;
for (i2 = 0; i2 < 12; i2++) {
f0 += fv1[i0 + 10 * i2] * xp1[i2 + 12 * i1];
}
fv0[i0 + 10 * i1] = (float)dv3[i0 + 10 * i1] + f0;
}
}
tansig_apply(fv0, fv2);
for (i0 = 0; i0 < 5; i0++) {
for (i1 = 0; i1 < 124; i1++) {
f0 = 0.0F;
for (i2 = 0; i2 < 10; i2++) {
f0 += fv4[i0 + 5 * i2] * fv2[i2 + 10 * i1];
}
fv3[i0 + 5 * i1] = (float)dv2[i0 + 5 * i1] + f0;
}
}
/* Output 1 */
softmax_apply(fv3, fv5);
mapminmax_reverse(fv5, b_y1);
}
/*
* function [lp]=gaussmixp(y,m,v,w)
* y = cat(1, testSamples(1).mfcc{:});
* m = gmm.M;
* v = gmm.V;
* w = gmm.W;
*/
void gaussmixp(const real_T y[2004], const real_T m[108], const real_T v[108],
const real_T w[9], real_T lp[167])
{
real_T b[108];
real_T dv0[108];
real_T b_b[9];
real_T lvm[9];
int32_T ix;
real_T mx[167];
real_T kk[1503];
real_T km[1503];
static real_T b_y[18036];
int32_T iy;
real_T dv1[18036];
real_T x[1503];
int32_T i;
int32_T ixstart;
int32_T ixstop;
real_T mtmp;
int32_T b_ix;
boolean_T exitg1;
real_T ps[167];
/* GAUSSMIXP calculate probability densities from a Gaussian mixture model */
/* */
/* Inputs: n data values, k mixtures, p parameters, q data vector size */
/* */
/* Y(n,q) = input data */
/* M(k,p) = mixture means for x(p) */
/* V(k,p) or V(p,p,k) variances (diagonal or full) */
/* W(k,1) = weights */
/* A(q,p), B(q) = transformation: y=x*a'+b' (where y and x are row vectors) */
/* if A is omitted, it is assumed to be the first q rows of the */
/* identity matrix. B defaults to zero. */
/* Note that most commonly, q=p and A and B are omitted entirely. */
/* */
/* Outputs */
/* */
/* LP(n,1) = log probability of each data point */
/* RP(n,k) = relative probability of each mixture */
/* KH(n,1) = highest probability mixture */
/* KP(n,1) = relative probability of highest probability mixture */
/* Copyright (C) Mike Brookes 2000-2009 */
/* Version: $Id: gaussmixp.m,v 1.3 2009/04/08 07:51:21 dmb Exp $ */
/* */
/* VOICEBOX is a MATLAB toolbox for speech processing. */
/* Home page: http://www.ee.ic.ac.uk/hp/staff/dmb/voicebox/voicebox.html */
/* */
/* %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% */
/* This program is free software; you can redistribute it and/or modify */
/* it under the terms of the GNU General Public License as published by */
/* the Free Software Foundation; either version 2 of the License, or */
/* (at your option) any later version. */
/* */
/* This program is distributed in the hope that it will be useful, */
/* but WITHOUT ANY WARRANTY; without even the implied warranty of */
/* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the */
/* GNU General Public License for more details. */
/* */
/* You can obtain a copy of the GNU General Public License from */
/* http://www.gnu.org/copyleft/gpl.html or by writing to */
/* Free Software Foundation, Inc.,675 Mass Ave, Cambridge, MA 02139, USA. */
/* %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% */
/* 'gaussmixp:49' [n,q]=size(y); */
/* 'gaussmixp:50' [k,p]=size(m); */
/* 'gaussmixp:51' memsize=voicebox('memsize'); */
voicebox();
/* set memory size to use */
/* 'gaussmixp:53' lp=zeros(n,1); */
memset((void *)&lp[0], 0, 167U * sizeof(real_T));
/* 'gaussmixp:54' wk=ones(k,1); */
/* 'gaussmixp:56' vi=-0.5*v.^(-1); */
power(v, b);
/* data-independent scale factor in exponent */
/* 'gaussmixp:57' lvm=log(w)-0.5*sum(log(v),2); */
memcpy((void *)&dv0[0], (void *)&v[0], 108U * sizeof(real_T));
c_log(dv0);
sum(dv0, b_b);
memcpy((void *)&lvm[0], (void *)&w[0], 9U * sizeof(real_T));
b_log(lvm);
for (ix = 0; ix < 9; ix++) {
lvm[ix] -= 0.5 * b_b[ix];
}
/* log of external scale factor (excluding -0.5*q*log(2pi) term) */
/* 'gaussmixp:58' ii=1:n; */
/* 'gaussmixp:59' wnj=ones(1,n); */
/* 'gaussmixp:60' kk=repmat(ii,k,1); */
for (ix = 0; ix < 167; ix++) {
mx[ix] = 1.0 + (real_T)ix;
}
repmat(mx, kk);
/* 'gaussmixp:61' km=repmat(1:k,1,n); */
for (ix = 0; ix < 9; ix++) {
b_b[ix] = 1.0 + (real_T)ix;
}
b_repmat(b_b, km);
/* 'gaussmixp:62' py=reshape(sum((y(kk(:),:)-m(km(:),:)).^2.*vi(km(:),:),2),k,n)+lvm(:,wnj); */
for (ix = 0; ix < 12; ix++) {
for (iy = 0; iy < 1503; iy++) {
b_y[iy + 1503 * ix] = y[((int32_T)kk[iy] + 167 * ix) - 1] - m[((int32_T)
km[iy] + 9 * ix) - 1];
}
}
b_power(b_y, dv1);
for (ix = 0; ix < 12; ix++) {
for (iy = 0; iy < 1503; iy++) {
b_y[iy + 1503 * ix] = dv1[iy + 1503 * ix] * (-0.5 * b[((int32_T)km[iy] + 9
* ix) - 1]);
}
}
b_sum(b_y, x);
memcpy((void *)&kk[0], (void *)&x[0], 1503U * sizeof(real_T));
for (ix = 0; ix < 167; ix++) {
for (iy = 0; iy < 9; iy++) {
km[iy + 9 * ix] = kk[iy + 9 * ix] + lvm[iy];
}
}
/* 'gaussmixp:63' mx=max(py,[],1); */
ix = -8;
iy = -1;
for (i = 0; i < 167; i++) {
ix += 9;
ixstart = ix;
ixstop = ix + 8;
mtmp = km[ix - 1];
if (rtIsNaN(km[ix - 1])) {
b_ix = ix;
exitg1 = 0U;
while ((exitg1 == 0U) && (b_ix + 1 <= ixstop)) {
ixstart = b_ix + 1;
if (!rtIsNaN(km[b_ix])) {
mtmp = km[b_ix];
exitg1 = 1U;
} else {
b_ix++;
}
}
}
if (ixstart < ixstop) {
while (ixstart + 1 <= ixstop) {
if (km[ixstart] > mtmp) {
mtmp = km[ixstart];
}
ixstart++;
}
}
iy++;
mx[iy] = mtmp;
}
/* find normalizing factor for each data point to prevent underflow when using exp() */
/* 'gaussmixp:64' px=exp(py-mx(wk,:)); */
for (ix = 0; ix < 167; ix++) {
for (iy = 0; iy < 9; iy++) {
kk[iy + 9 * ix] = km[iy + 9 * ix] - mx[ix];
}
}
b_exp(kk);
/* find normalized probability of each mixture for each datapoint */
/* 'gaussmixp:65' ps=sum(px,1); */
c_sum(kk, ps);
/* total normalized likelihood of each data point */
/* 'gaussmixp:66' lp(ii)=log(ps)+mx; */
d_log(ps);
for (ix = 0; ix < 167; ix++) {
/* 'gaussmixp:67' lp=lp-0.5*q*log(2*pi); */
lp[ix] = (ps[ix] + mx[ix]) - 11.027262398456072;
}
}
//
// Arguments : const emxArray_uint8_T *a
// signed char direction
// emxArray_uint8_T *b
// Return Type : void
//
void b_ConstantPad(const emxArray_uint8_T *a, signed char direction,
emxArray_uint8_T *b)
{
int sizeB[2];
int cdiff;
int i37;
emxArray_uint8_T *r14;
int b_sizeB[2];
int absb;
int ndbl;
emxArray_real_T *y;
emxArray_real_T *b_y;
int apnd;
for (cdiff = 0; cdiff < 2; cdiff++) {
if (direction == 5) {
i37 = a->size[cdiff] + 1;
sizeB[cdiff] = i37;
} else if (direction == 6) {
i37 = a->size[cdiff] + 1;
sizeB[cdiff] = i37;
} else {
i37 = a->size[cdiff];
sizeB[cdiff] = (int)((double)i37 + 2.0);
}
}
b_emxInit_uint8_T(&r14, 2);
b_sizeB[0] = sizeB[0];
b_sizeB[1] = sizeB[1];
b_repmat(b_sizeB, r14);
for (i37 = 0; i37 < 2; i37++) {
absb = b->size[0] * b->size[1];
b->size[i37] = r14->size[i37];
emxEnsureCapacity((emxArray__common *)b, absb, (int)sizeof(unsigned char));
}
if (direction == 5) {
for (cdiff = 0; cdiff < r14->size[0]; cdiff++) {
b->data[cdiff] = 0;
}
i37 = b->size[1];
for (ndbl = 2; ndbl <= i37; ndbl++) {
b->data[b->size[0] * (ndbl - 1)] = 0;
}
} else if (direction == 7) {
for (cdiff = 0; cdiff < r14->size[0]; cdiff++) {
b->data[cdiff] = 0;
}
i37 = b->size[1];
for (ndbl = a->size[1] + 1; ndbl + 1 <= i37; ndbl++) {
absb = b->size[0];
for (cdiff = 0; cdiff < absb; cdiff++) {
b->data[cdiff + b->size[0] * ndbl] = 0;
}
}
for (ndbl = 2; ndbl <= a->size[1] + 1; ndbl++) {
b->data[b->size[0] * (ndbl - 1)] = 0;
}
for (ndbl = 2; ndbl <= a->size[1] + 1; ndbl++) {
i37 = b->size[0];
for (cdiff = a->size[0] + 1; cdiff + 1 <= i37; cdiff++) {
b->data[cdiff + b->size[0] * (ndbl - 1)] = 0;
}
}
} else {
for (ndbl = a->size[1]; ndbl + 1 <= r14->size[1]; ndbl++) {
i37 = b->size[0];
for (cdiff = 0; cdiff < i37; cdiff++) {
b->data[cdiff + b->size[0] * ndbl] = 0;
}
}
for (ndbl = 0; ndbl < a->size[1]; ndbl++) {
i37 = b->size[0];
for (cdiff = a->size[0]; cdiff + 1 <= i37; cdiff++) {
b->data[cdiff + b->size[0] * ndbl] = 0;
}
}
}
emxFree_uint8_T(&r14);
b_emxInit_real_T(&y, 2);
b_emxInit_real_T(&b_y, 2);
if ((direction == 5) || (direction == 7)) {
if (a->size[0] < 1) {
absb = -1;
apnd = 0;
} else {
ndbl = (int)floor(((double)a->size[0] - 1.0) + 0.5);
apnd = ndbl + 1;
cdiff = (ndbl - a->size[0]) + 1;
absb = a->size[0];
if (fabs((double)cdiff) < 4.4408920985006262E-16 * (double)absb) {
ndbl++;
apnd = a->size[0];
} else if (cdiff > 0) {
apnd = ndbl;
} else {
ndbl++;
}
absb = ndbl - 1;
}
i37 = y->size[0] * y->size[1];
y->size[0] = 1;
y->size[1] = absb + 1;
emxEnsureCapacity((emxArray__common *)y, i37, (int)sizeof(double));
if (absb + 1 > 0) {
y->data[0] = 1.0;
if (absb + 1 > 1) {
y->data[absb] = apnd;
ndbl = absb / 2;
for (cdiff = 1; cdiff < ndbl; cdiff++) {
y->data[cdiff] = 1.0 + (double)cdiff;
y->data[absb - cdiff] = apnd - cdiff;
}
if (ndbl << 1 == absb) {
y->data[ndbl] = (1.0 + (double)apnd) / 2.0;
} else {
y->data[ndbl] = 1.0 + (double)ndbl;
y->data[ndbl + 1] = apnd - ndbl;
}
}
}
if (a->size[1] < 1) {
absb = -1;
apnd = 0;
} else {
ndbl = (int)floor(((double)a->size[1] - 1.0) + 0.5);
apnd = ndbl + 1;
cdiff = (ndbl - a->size[1]) + 1;
absb = a->size[1];
if (fabs((double)cdiff) < 4.4408920985006262E-16 * (double)absb) {
ndbl++;
apnd = a->size[1];
} else if (cdiff > 0) {
apnd = ndbl;
} else {
ndbl++;
}
absb = ndbl - 1;
}
i37 = b_y->size[0] * b_y->size[1];
b_y->size[0] = 1;
b_y->size[1] = absb + 1;
emxEnsureCapacity((emxArray__common *)b_y, i37, (int)sizeof(double));
if (absb + 1 > 0) {
b_y->data[0] = 1.0;
if (absb + 1 > 1) {
b_y->data[absb] = apnd;
ndbl = absb / 2;
for (cdiff = 1; cdiff < ndbl; cdiff++) {
b_y->data[cdiff] = 1.0 + (double)cdiff;
b_y->data[absb - cdiff] = apnd - cdiff;
}
if (ndbl << 1 == absb) {
b_y->data[ndbl] = (1.0 + (double)apnd) / 2.0;
} else {
b_y->data[ndbl] = 1.0 + (double)ndbl;
b_y->data[ndbl + 1] = apnd - ndbl;
}
}
}
ndbl = a->size[1];
for (i37 = 0; i37 < ndbl; i37++) {
cdiff = a->size[0];
for (absb = 0; absb < cdiff; absb++) {
b->data[(int)y->data[y->size[0] * absb] + b->size[0] * (int)b_y->
data[b_y->size[0] * i37]] = a->data[absb + a->size[0] * i37];
}
}
} else {
ndbl = a->size[1];
for (i37 = 0; i37 < ndbl; i37++) {
cdiff = a->size[0];
for (absb = 0; absb < cdiff; absb++) {
b->data[absb + b->size[0] * i37] = a->data[absb + a->size[0] * i37];
}
}
}
emxFree_real_T(&b_y);
emxFree_real_T(&y);
}
