void vad(float rc,
float * lsf,
float * rxx,
float * sigpp,
int frm_count,
int prev_marker, int pprev_marker, int *marker, float * Energy_db)
{
float tmp[M];
float SD;
float E_low;
float dtemp;
float dSE;
float dSLE;
float ZC;
float COEF;
float COEFZC;
float COEFSD;
float dSZC;
float norm_energy;
int i;
/* compute the frame energy */
norm_energy =
(float) 10.0 *(float) log10((float) (rxx[0] / (float) 240.0 + EPSI));
*Energy_db = norm_energy;
/* compute the low band energy */
E_low = (float) 0.0;
for (i = 1; i <= NP; i++)
E_low = E_low + rxx[i] * lbf_corr[i];
E_low = rxx[0] * lbf_corr[0] + (float) 2.0 *E_low;
if (E_low < (float) 0.0)
E_low = (float) 0.0;
E_low = (float) 10.0 *(float) log10((float) (E_low / (float) 240.0 + EPSI));
/* compute SD */
/* Normalize lsfs */
for (i = 0; i < M; i++)
lsf[i] /= (float) 2. *PI;
dvsub(lsf, MeanLSF, tmp, M);
SD = dvdot(tmp, tmp, M);
/* compute # zero crossing */
ZC = (float) 0.0f;
dtemp = sigpp[ZC_START];
for (i = ZC_START + 1; i <= ZC_END; i++) {
if (dtemp * sigpp[i] < (float) 0.0) {
ZC = ZC + (float) 1.0;
}
dtemp = sigpp[i];
}
ZC = ZC / (float) 80.0;
/* Initialize and update Mins */
if (frm_count < 129) {
if (norm_energy < Min) {
Min = norm_energy;
Prev_Min = norm_energy;
}
if ((frm_count % 8) == 0) {
Min_buffer[(int)frm_count / 8 - 1] = Min;
Min = FLT_MAX_G729;
}
}
if ((frm_count % 8) == 0) {
Prev_Min = Min_buffer[0];
for (i = 1; i < 15; i++) {
if (Min_buffer[i] < Prev_Min)
Prev_Min = Min_buffer[i];
}
}
if (frm_count >= 129) {
if ((frm_count % 8) == 1) {
Min = Prev_Min;
Next_Min = FLT_MAX_G729;
}
if (norm_energy < Min)
Min = norm_energy;
if (norm_energy < Next_Min)
Next_Min = norm_energy;
if ((frm_count % 8) == 0) {
for (i = 0; i < 15; i++)
Min_buffer[i] = Min_buffer[i + 1];
Min_buffer[15] = Next_Min;
Prev_Min = Min_buffer[0];
for (i = 1; i < 16; i++) {
if (Min_buffer[i] < Prev_Min)
Prev_Min = Min_buffer[i];
}
}
}
if (frm_count <= INIT_FRAME) {
if (norm_energy < (float) 21.0) {
less_count++;
*marker = NOISE;
} else {
*marker = VOICE;
MeanE =
(MeanE * ((float) (frm_count - less_count - 1)) +
norm_energy) / (float) (frm_count - less_count);
MeanSZC =
(MeanSZC * ((float) (frm_count - less_count - 1)) +
ZC) / (float) (frm_count - less_count);
dvwadd(MeanLSF, (float) (frm_count - less_count - 1),
lsf, (float) 1.0, MeanLSF, M);
dvsmul(MeanLSF,
(float) 1.0 / (float) (frm_count - less_count),
MeanLSF, M);
}
}
if (frm_count >= INIT_FRAME) {
if (frm_count == INIT_FRAME) {
MeanSE = MeanE - (float) 10.0;
MeanSLE = MeanE - (float) 12.0;
}
dSE = MeanSE - norm_energy;
dSLE = MeanSLE - E_low;
dSZC = MeanSZC - ZC;
if (norm_energy < (float) 21.0) {
*marker = NOISE;
} else {
*marker = MakeDec(dSLE, dSE, SD, dSZC);
}
v_flag = 0;
if ((prev_marker == VOICE) && (*marker == NOISE) &&
(norm_energy > MeanSE + (float) 2.0)
&& (norm_energy > (float) 21.0)) {
*marker = VOICE;
v_flag = 1;
}
if (flag == 1) {
if ((pprev_marker == VOICE) && (prev_marker == VOICE) &&
(*marker == NOISE)
&& (fabs(prev_energy - norm_energy) <= (float) 3.0)) {
count_ext++;
*marker = VOICE;
v_flag = 1;
if (count_ext <= 4)
flag = 1;
else {
flag = 0;
count_ext = 0;
}
}
} else
flag = 1;
if (*marker == NOISE)
count_sil++;
if ((*marker == VOICE) && (count_sil > 10) &&
((norm_energy - prev_energy) <= (float) 3.0)) {
*marker = NOISE;
count_sil = 0;
}
if (*marker == VOICE)
count_sil = 0;
if ((norm_energy < MeanSE + (float) 3.0) && (frm_count > 128)
&& (!v_flag) && (rc < (float) 0.6))
*marker = NOISE;
if ((norm_energy < MeanSE + (float) 3.0) && (rc < (float) 0.75)
&& (SD < (float) 0.002532959)) {
count_update++;
if (count_update < INIT_COUNT) {
COEF = (float) 0.75;
COEFZC = (float) 0.8;
COEFSD = (float) 0.6;
} else if (count_update < INIT_COUNT + 10) {
COEF = (float) 0.95;
COEFZC = (float) 0.92;
COEFSD = (float) 0.65;
} else if (count_update < INIT_COUNT + 20) {
COEF = (float) 0.97;
COEFZC = (float) 0.94;
COEFSD = (float) 0.70;
} else if (count_update < INIT_COUNT + 30) {
COEF = (float) 0.99;
COEFZC = (float) 0.96;
COEFSD = (float) 0.75;
} else if (count_update < INIT_COUNT + 40) {
COEF = (float) 0.995;
COEFZC = (float) 0.99;
COEFSD = (float) 0.75;
} else {
COEF = (float) 0.995;
COEFZC = (float) 0.998;
COEFSD = (float) 0.75;
}
dvwadd(MeanLSF, COEFSD, lsf, (float) 1.0 - COEFSD, MeanLSF,
M);
MeanSE = COEF * MeanSE + ((float) 1.0 - COEF) * norm_energy;
MeanSLE = COEF * MeanSLE + ((float) 1.0 - COEF) * E_low;
MeanSZC = COEFZC * MeanSZC + ((float) 1.0 - COEFZC) * ZC;
}
if (((frm_count > 128)
&& ((MeanSE < Min) && (SD < (float) 0.002532959)))
|| (MeanSE > Min + (float) 10.0)) {
MeanSE = Min;
count_update = 0;
}
}
prev_energy = norm_energy;
return;
}
void vad(struct vad_state_t * state,
GFLOAT rc,
GFLOAT *lsf,
GFLOAT *rxx,
GFLOAT *sigpp,
int frm_count,
int prev_marker,
int pprev_marker,
int *marker,
GFLOAT *Energy_db)
{
GFLOAT tmp[M];
GFLOAT SD;
GFLOAT E_low;
GFLOAT dtemp;
GFLOAT dSE;
GFLOAT dSLE;
GFLOAT ZC;
GFLOAT COEF;
GFLOAT COEFZC;
GFLOAT COEFSD;
GFLOAT dSZC;
GFLOAT norm_energy;
int i;
/* compute the frame energy */
norm_energy = (F)10.0*(GFLOAT) log10((GFLOAT)( rxx[0]/(F)240.0 +EPSI));
*Energy_db = norm_energy ;
/* compute the low band energy */
E_low = (F)0.0;
for( i=1; i<= NP; i++)
E_low = E_low + rxx[i]*lbf_corr[i];
E_low= rxx[0]*lbf_corr[0] + (F)2.0*E_low;
if (E_low < (F)0.0) E_low = (F)0.0;
E_low= (F)10.0*(GFLOAT) log10((GFLOAT) (E_low/(F)240.0+EPSI));
/* compute SD */
/* Normalize lsfs */
for(i=0; i<M; i++) lsf[i] /= (F)2.*PI;
dvsub(lsf, state->MeanLSF,tmp,M);
SD = dvdot(tmp,tmp,M);
/* compute # zero crossing */
ZC = (F)0.0f;
dtemp = sigpp[ZC_START];
for (i=ZC_START+1 ; i <= ZC_END ; i++) {
if (dtemp*sigpp[i] < (F)0.0) {
ZC= ZC +(F)1.0;
}
dtemp = sigpp[i];
}
ZC = ZC/(F)80.0;
/* Initialize and update Mins */
if( frm_count < 129 ) {
if( norm_energy < state->Min ){
state->Min = norm_energy;
state->Prev_Min = norm_energy;
}
if( (frm_count % 8) == 0){
state->Min_buffer[(int)frm_count/8 -1] = state->Min;
state->Min = FLT_MAX_G729;
}
}
if( (frm_count % 8) == 0){
state->Prev_Min = state->Min_buffer[0];
for ( i =1; i< 15; i++){
if ( state->Min_buffer[i] < state->Prev_Min )
state->Prev_Min = state->Min_buffer[i];
}
}
if( frm_count >= 129 ) {
if( (frm_count % 8 ) == 1) {
state->Min = state->Prev_Min;
state->Next_Min = FLT_MAX_G729;
}
if( norm_energy < state->Min )
state->Min = norm_energy;
if( norm_energy < state->Next_Min )
state->Next_Min = norm_energy;
if( (frm_count % 8) == 0){
for ( i =0; i< 15; i++)
state->Min_buffer[i] = state->Min_buffer[i+1];
state->Min_buffer[15] = state->Next_Min;
state->Prev_Min = state->Min_buffer[0];
for ( i =1; i< 16; i++){
if ( state->Min_buffer[i] < state->Prev_Min )
state->Prev_Min = state->Min_buffer[i];
}
}
}
if (frm_count <= INIT_FRAME){
if( norm_energy < (F)21.0){
state->less_count++;
*marker = NOISE;
}
else{
*marker = VOICE;
state->MeanE = (state->MeanE * ( (GFLOAT)(frm_count - state->less_count -1)) +
norm_energy)/(GFLOAT) (frm_count - state->less_count);
state->MeanSZC = (state->MeanSZC * ( (GFLOAT)(frm_count - state->less_count -1)) +
ZC)/(GFLOAT) (frm_count - state->less_count);
dvwadd(state->MeanLSF,(GFLOAT) (frm_count - state->less_count -1),lsf,(F)1.0,state->MeanLSF,M);
dvsmul(state->MeanLSF,(F)1.0/(GFLOAT) (frm_count - state->less_count ), state->MeanLSF,M);
}
}
if (frm_count >= INIT_FRAME ){
if (frm_count == INIT_FRAME ){
state->MeanSE = state->MeanE -(F)10.0;
state->MeanSLE = state->MeanE -(F)12.0;
}
dSE = state->MeanSE - norm_energy;
dSLE = state->MeanSLE - E_low;
dSZC = state->MeanSZC - ZC;
if( norm_energy < (F)21.0 ){
*marker = NOISE;
}
else{
*marker = vad_make_dec(dSLE, dSE, SD, dSZC );
}
state->v_flag =0;
if( (prev_marker == VOICE) && (*marker == NOISE) &&
(norm_energy > state->MeanSE + (F)2.0) && ( norm_energy>(F)21.0)){
*marker = VOICE;
state->v_flag=1;
}
if((state->flag == 1) ){
if( (pprev_marker == VOICE) && (prev_marker == VOICE) &&
(*marker == NOISE) && (fabs(state->prev_energy - norm_energy)<= (F)3.0)){
state->count_ext++;
*marker = VOICE;
state->v_flag=1;
if(state->count_ext <=4)
state->flag =1;
else{
state->flag =0;
state->count_ext=0;
}
}
}
else
state->flag =1;
if(*marker == NOISE)
state->count_sil++;
if((*marker == VOICE) && (state->count_sil > 10) &&
((norm_energy - state->prev_energy) <= (F)3.0)){
*marker = NOISE;
state->count_sil=0;
}
if(*marker == VOICE)
state->count_sil=0;
if ((norm_energy < state->MeanSE + (F)3.0) && ( frm_count >128) &&( !state->v_flag) && (rc <(F)0.6) )
*marker = NOISE;
if ((norm_energy < state->MeanSE + (F)3.0) && (rc <(F)0.75) && ( SD<(F)0.002532959)){
state->count_update++;
if (state->count_update < INIT_COUNT){
COEF = (F)0.75;
COEFZC = (F)0.8;
COEFSD = (F)0.6;
}
else
if (state->count_update < INIT_COUNT+10){
COEF = (F)0.95;
COEFZC = (F)0.92;
COEFSD = (F)0.65;
}
else
if (state->count_update < INIT_COUNT+20){
COEF = (F)0.97;
COEFZC = (F)0.94;
COEFSD = (F)0.70;
}
else
if (state->count_update < INIT_COUNT+30){
COEF = (F)0.99;
COEFZC = (F)0.96;
COEFSD = (F)0.75;
}
else
if (state->count_update < INIT_COUNT+40){
COEF = (F)0.995;
COEFZC = (F)0.99;
COEFSD = (F)0.75;
}
else{
COEF = (F)0.995;
COEFZC = (F)0.998;
COEFSD = (F)0.75;
}
dvwadd(state->MeanLSF,COEFSD,lsf,(F)1.0 -COEFSD,state->MeanLSF,M);
state->MeanSE = COEF * state->MeanSE+((F)1.0- COEF)*norm_energy;
state->MeanSLE = COEF * state->MeanSLE+((F)1.0- COEF)*E_low;
state->MeanSZC = COEFZC * state->MeanSZC+((F)1.0- COEFZC)*ZC;
}
if(((frm_count > 128) && (state->MeanSE < state->Min) && (SD < (F)0.002532959)) || (state->MeanSE > state->Min + (F)10.0) )
{
state->MeanSE = state->Min;
state->count_update = 0;
}
}
state->prev_energy = norm_energy;
return;
}
