www.pudn.com > p_voice.rar > PITCHCP.C
/*
ITU-T G.729 Annex C+ - Reference C code for floating point
implementation of G.729 Annex C+
(integration of Annexes B, D and E)
Version 2.1 of October 1999
*/
/*
File : PITCHCP.C
*/
/*****************************************************************************/
/* Long Term Prediction Routines */
/*****************************************************************************/
#include
#include "typedef.h"
#include "ld8k.h"
#include "tab_ld8k.h"
#include "ld8cp.h"
/* prototypes for local functions */
static void norm_corr(FLOAT exc[], FLOAT xn[], FLOAT h[], int l_subfr,
int t_min, int t_max, FLOAT corr_norm[]);
static FLOAT interpol_3(FLOAT cor[], int frac);
static int lag_max(FLOAT signal[],int L_frame,int pit_max,
int pit_min, FLOAT *cor_max);
static FLOAT inv_sqrt( FLOAT x );
/*----------------------------------------------------------------------------
* pitch_ol - compute the open loop pitch lag
*----------------------------------------------------------------------------
*/
int pitch_ol( /* output: open-loop pitch lag */
FLOAT signal[], /* input : signal to compute pitch */
/* s[-PIT_MAX : l_frame-1] */
int pit_min, /* input : minimum pitch lag */
int pit_max, /* input : maximum pitch lag */
int l_frame /* input : error minimization window */
)
{
FLOAT max1, max2, max3;
int p_max1, p_max2, p_max3;
/*--------------------------------------------------------------------*
* The pitch lag search is divided in three sections. *
* Each section cannot have a pitch multiple. *
* We find a maximum for each section. *
* We compare the maxima of each section by favoring small lag. *
* *
* First section: lag delay = PIT_MAX to 80 *
* Second section: lag delay = 79 to 40 *
* Third section: lag delay = 39 to 20 *
*--------------------------------------------------------------------*/
p_max1 = lag_max(signal, l_frame, pit_max, 80 , &max1);
p_max2 = lag_max(signal, l_frame, 79 , 40 , &max2);
p_max3 = lag_max(signal, l_frame, 39 , pit_min , &max3);
/*--------------------------------------------------------------------*
* Compare the 3 sections maxima, and favor small lag. *
*--------------------------------------------------------------------*/
if ( max1 * THRESHPIT < max2 ) {
max1 = max2;
p_max1 = p_max2;
}
if ( max1 * THRESHPIT < max3 ) p_max1 = p_max3;
return (p_max1);
}
/*----------------------------------------------------------------------------
* lag_max - Find the lag that has maximum correlation
*----------------------------------------------------------------------------
*/
static int lag_max( /* output: lag found */
FLOAT signal[], /* input : Signal to compute the open loop pitch
signal[-142:-1] should be known. */
int l_frame, /* input : Length of frame to compute pitch */
int lagmax, /* input : maximum lag */
int lagmin, /* input : minimum lag */
FLOAT *cor_max /* input : normalized correlation of selected lag */
)
{
int i, j;
FLOAT *p, *p1;
FLOAT max, t0;
int p_max;
max = FLT_MIN_G729;
p_max = lagmax; /* to avoid visual warning */
for (i = lagmax; i >= lagmin; i--) {
p = signal;
p1 = &signal[-i];
t0 = (F)0.0;
for (j=0; j= max) {
max = t0;
p_max = i;
}
}
/* compute energy */
t0 = (F)0.01; /* to avoid division by zero */
p = &signal[-p_max];
for(i=0; i= max)
{
max = corr[i];
lag = i;
}
}
/* If first subframe and lag > 84 do not search fractionnal pitch */
if( (i_subfr == 0) && (lag > 84) )
{
*pit_frac = 0;
return(lag);
}
/* test the fractions around lag and choose the one which maximizes
the interpolated normalized correlation */
if (rate == G729D) { /* 6.4 kbps */
if (i_subfr == 0) {
max = interpol_3(&corr[lag], -2);
frac = -2;
for (i = -1; i <= 2; i++) {
corr_int = interpol_3(&corr[lag], i);
if(corr_int > max) {
max = corr_int;
frac = i;
}
}
}
else {
midLag = t0_max - 4;
if ((lag == midLag - 1) || lag == midLag) {
max = interpol_3(&corr[lag], -2);
frac = -2;
for (i = -1; i <= 2; i++) {
corr_int = interpol_3(&corr[lag], i);
if(corr_int > max) {
max = corr_int;
frac = i;
}
}
}
else if (lag == midLag - 2) {
max = interpol_3(&corr[lag], 0);
frac = 0;
for (i = 1; i <= 2; i++) {
corr_int = interpol_3(&corr[lag], i);
if(corr_int > max) {
max = corr_int;
frac = i;
}
}
}
else if (lag == midLag + 1) {
max = interpol_3(&corr[lag], -2);
frac = -2;
for (i = -1; i <= 0; i++) {
corr_int = interpol_3(&corr[lag], i);
if(corr_int > max) {
max = corr_int;
frac = i;
}
}
}
else
frac = 0;
}
}
else {
max = interpol_3(&corr[lag], -2);
frac = -2;
for (i = -1; i <= 2; i++)
{
corr_int = interpol_3(&corr[lag], i);
if(corr_int > max)
{
max = corr_int;
frac = i;
}
}
}
/* limit the fraction value in the interval [-1,0,1] */
if (frac == -2)
{
frac = 1;
lag -= 1;
}
if (frac == 2)
{
frac = -1;
lag += 1;
}
*pit_frac = frac;
return lag;
}
/*----------------------------------------------------------------------------
* norm_corr - Find the normalized correlation between the target vector and
* the filtered past excitation.
*----------------------------------------------------------------------------
*/
static void norm_corr(
FLOAT exc[], /* input : excitation buffer */
FLOAT xn[], /* input : target vector */
FLOAT h[], /* input : imp response of synth and weighting flt */
int l_subfr, /* input : Length of frame to compute pitch */
int t_min, /* input : minimum value of searched range */
int t_max, /* input : maximum value of search range */
FLOAT corr_norm[] /* output: normalized correlation (correlation
between target and filtered excitation
divided by the square root of energy of
filtered excitation) */
)
{
int i, j, k;
FLOAT excf[L_SUBFR]; /* filtered past excitation */
FLOAT alp, s, norm;
k = -t_min;
/* compute the filtered excitation for the first delay t_min */
convolve(&exc[k], h, excf, l_subfr);
/* loop for every possible period */
for (i = t_min; i <= t_max; i++)
{
/* Compute 1/sqrt(energie of excf[]) */
alp = (F)0.01;
for (j = 0; j < l_subfr; j++)
alp += excf[j]*excf[j];
norm = inv_sqrt(alp);
/* Compute correlation between xn[] and excf[] */
s = (F)0.0;
for (j = 0; j < l_subfr; j++) s += xn[j]*excf[j];
/* Normalize correlation = correlation * (1/sqrt(energie)) */
corr_norm[i] = s*norm;
/* modify the filtered excitation excf[] for the next iteration */
if (i != t_max)
{
k--;
for (j = l_subfr-1; j > 0; j--)
excf[j] = excf[j-1] + exc[k]*h[j];
excf[0] = exc[k];
}
}
return;
}
/*----------------------------------------------------------------------------
* g_pitch - compute adaptive codebook gain and compute , -2
*----------------------------------------------------------------------------
*/
FLOAT g_pitch( /* output: pitch gain */
FLOAT xn[], /* input : target vector */
FLOAT y1[], /* input : filtered adaptive codebook vector */
FLOAT g_coeff[], /* output: and -2 */
int l_subfr /* input : vector dimension */
)
{
FLOAT xy, yy, gain;
int i;
xy = (F)0.0;
for (i = 0; i < l_subfr; i++) {
xy += xn[i] * y1[i];
}
yy = (F)0.01;
for (i = 0; i < l_subfr; i++) {
yy += y1[i] * y1[i]; /* energy of filtered excitation */
}
g_coeff[0] = yy;
g_coeff[1] = (F)-2.0*xy +(F)0.01;
/* find pitch gain and bound it by [0,1.2] */
gain = xy/yy;
if (gain<(F)0.0) gain = (F)0.0;
if (gain>GAIN_PIT_MAX) gain = GAIN_PIT_MAX;
return gain;
}
/*----------------------------------------------------------------------*
* Function enc_lag3cp() *
* ~~~~~~~~~~ *
* Encoding of fractional pitch lag with 1/3 resolution. *
*----------------------------------------------------------------------*
* The pitch range for the first subframe is divided as follows: *
* 19 1/3 to 84 2/3 resolution 1/3 *
* 85 to 143 resolution 1 *
* *
* The period in the first subframe is encoded with 8 bits. *
* For the range with fractions: *
* index = (T-19)*3 + frac - 1; where T=[19..85] and frac=[-1,0,1] *
* and for the integer only range *
* index = (T - 85) + 197; where T=[86..143] *
*----------------------------------------------------------------------*
* For the second subframe a resolution of 1/3 is always used, and the *
* search range is relative to the lag in the first subframe. *
* If t0 is the lag in the first subframe then *
* t_min=t0-5 and t_max=t0+4 and the range is given by *
* t_min - 2/3 to t_max + 2/3 *
* *
* The period in the 2nd subframe is encoded with 5 bits: *
* index = (T-(t_min-1))*3 + frac - 1; where T[t_min-1 .. t_max+1] *
*----------------------------------------------------------------------*/
int enc_lag3cp( /* output: Return index of encoding */
int T0, /* input : Pitch delay */
int T0_frac, /* input : Fractional pitch delay */
int *T0_min, /* in/out: Minimum search delay */
int *T0_max, /* in/out: Maximum search delay */
int pit_min, /* input : Minimum pitch delay */
int pit_max, /* input : Maximum pitch delay */
int pit_flag, /* input : Flag for 1st subframe */
int rate
)
{
int index;
if (pit_flag == 0) /* if 1st subframe */
{
/* encode pitch delay (with fraction) */
if (T0 <= 85)
index = T0*3 - 58 + T0_frac;
else
index = T0 + 112;
/* find T0_min and T0_max for second subframe */
*T0_min = T0 - 5;
if (*T0_min < pit_min) *T0_min = pit_min;
*T0_max = *T0_min + 9;
if (*T0_max > pit_max)
{
*T0_max = pit_max;
*T0_min = *T0_max - 9;
}
}
else /* second subframe */
{
if (rate == G729D) { /* 4 bits in 2nd subframe (6.4 kbps) */
if (T0 < *T0_min + 3)
index = T0 - *T0_min;
else if (T0 < *T0_min + 7)
index = (T0 - (*T0_min + 3)) * 3 + T0_frac + 3;
else
index = (T0 - (*T0_min + 7)) + 13;
}
else {
index = T0 - *T0_min;
index = index*3 + 2 + T0_frac;
}
}
return index;
}
/*----------------------------------------------------------------------------
* interpol_3 - For interpolating the normalized correlation
*----------------------------------------------------------------------------
*/
static FLOAT interpol_3( /* output: interpolated value */
FLOAT *x, /* input : function to be interpolated */
int frac /* input : fraction value to evaluate */
)
{
int i;
FLOAT s, *x1, *x2, *c1, *c2;
if (frac < 0) {
frac += UP_SAMP;
x--;
}
x1 = &x[0];
x2 = &x[1];
c1 = &inter_3[frac];
c2 = &inter_3[UP_SAMP-frac];
s = (F)0.0;
for(i=0; i< L_INTER4; i++, c1+=UP_SAMP, c2+=UP_SAMP)
s+= (*x1--) * (*c1) + (*x2++) * (*c2);
return s;
}
/*----------------------------------------------------------------------------
* inv_sqrt - compute y = 1 / sqrt(x)
*----------------------------------------------------------------------------
*/
static FLOAT inv_sqrt( /* output: 1/sqrt(x) */
FLOAT x /* input : value of x */
)
{
return ((F)1.0 / (FLOAT)sqrt((double)x) );
}