Skip to content

GitLab

Commit bafbcd9b authored by Monty's avatar Monty
Browse files
Options 

Incremental update

svn path=/branches/unlabeled-1.16.2/vorbis/; revision=287
parent 536b2aa3
Hide whitespace changes
Inline Side-by-side
Showing with 104 additions and 148 deletions
lib/psy.c
View file @ bafbcd9b
......@@ -12,7 +12,7 @@
********************************************************************
function: psychoacoustics not including preecho
last mod: $Id: psy.c,v 1.16.2.1 2000/03/29 03:49:28 xiphmont Exp $
last mod: $Id: psy.c,v 1.16.2.2 2000/03/29 20:08:49 xiphmont Exp $
********************************************************************/
......@@ -26,6 +26,9 @@
#include "psy.h"
#include "scales.h"
/* the beginnings of real psychoacoustic infrastructure. This is
still not tightly tuned */
void _vi_psy_free(vorbis_info_psy *i){
if(i){
memset(i,0,sizeof(vorbis_info_psy));
......@@ -74,25 +77,26 @@ static void linear_curve(double *c){
c[i]=fromdB(c[i]);
}
static void interp_curve_dB(double *c,double *c1,double *c2,double del){
int i;
for(i=0;i<EHMER_MAX;i++)
c[i]=fromdB(todB(c2[i])*del+todB(c1[i])*(1.-del));
}
static void interp_curve(double *c,double *c1,double *c2,double del){
int i;
for(i=0;i<EHMER_MAX;i++)
c[i]=c2[i]*del+c1[i]*(1.-del);
}
extern void analysis(char *base,int i,double *v,int n,int bark,int dB);
static double c0[EHMER_MAX]; /* initialized to zero */
static void setup_curve(double **c,
int oc,
double dB20,
double dB40,
double dB60,
double dB80,
double dB100,
double dB120){
double *curveatt_dB,
double *peakatt_dB,
double peaklowrolloff,
double peakhighrolloff){
int i,j;
double tempc[11][EHMER_MAX];
double tempc[9][EHMER_MAX];
double ath[EHMER_MAX];
for(i=0;i<EHMER_MAX;i++){
......@@ -103,70 +107,66 @@ static void setup_curve(double **c,
ath[i]=ATH_Bark_dB[ibark]*(1.-del)+ATH_Bark_dB[ibark+1]*del;
else
ath[i]=200;
fprintf(stderr,"%d.%g ",ibark,del);
}
fprintf(stderr,"\n");
analysis("Cath",oc*100,ath,EHMER_MAX,0,0);
memcpy(c[10],c[8],sizeof(double)*EHMER_MAX);
memcpy(c[0],c[2],sizeof(double)*EHMER_MAX);
for(i=0;i<11;i+=2)
analysis("Cpre",oc*100+i,c[i],EHMER_MAX,0,0);
/* the temp curves are a bit roundabout, but this is only in
init. */
for(i=0;i<11;i++){
analysis("Cpre",oc*100+i,c[i],EHMER_MAX,0,0);
for(i=0;i<9;i++){
memcpy(tempc[i],c[i],sizeof(double)*EHMER_MAX);
max_curve(tempc[i],ath);
}
attenuate_curve(c[0],dB20);
attenuate_curve(c[2],dB40);
attenuate_curve(c[4],dB60);
attenuate_curve(c[6],dB80);
attenuate_curve(c[8],dB100);
attenuate_curve(c[10],dB120);
attenuate_curve(tempc[0],dB20);
attenuate_curve(tempc[2],dB40);
attenuate_curve(tempc[4],dB60);
attenuate_curve(tempc[6],dB80);
attenuate_curve(tempc[8],dB100);
attenuate_curve(tempc[10],dB120);
/* The c array is comes in as dB curves at 20, 40, 60 80 100 120 dB.
/* normalize them so the driving amplitude is 0dB */
for(i=0;i<5;i++){
attenuate_curve(c[i*2],curveatt_dB[i]);
attenuate_curve(tempc[i*2],curveatt_dB[i]);
}
/* add in the additional peak attenuation hack */
for(i=0;i<5;i++){
double att=peakatt_dB[i];
for(j=EHMER_OFFSET;j<EHMER_MAX && att>tempc[i*2][j];j++){
c[i*2][j]=att;
tempc[i*2][j]=att;
att+=peakhighrolloff;
}
att=peakatt_dB[i]+peaklowrolloff;
for(j=EHMER_OFFSET-1;j>=0 && att>tempc[i*2][j];j--){
c[i*2][j]=att;
tempc[i*2][j]=att;
att+=peaklowrolloff;
}
}
/* The c array is comes in as dB curves at 20 40 60 80 100 dB.
interpolate intermediate dB curves */
interp_curve(c[1],c[0],c[2],.5);
interp_curve(c[3],c[2],c[4],.5);
interp_curve(c[5],c[4],c[6],.5);
interp_curve(c[7],c[6],c[8],.5);
interp_curve(c[9],c[8],c[10],.5);
interp_curve(tempc[1],tempc[0],tempc[2],.5);
interp_curve(tempc[3],tempc[2],tempc[4],.5);
interp_curve(tempc[5],tempc[4],tempc[6],.5);
interp_curve(tempc[7],tempc[6],tempc[8],.5);
interp_curve(tempc[9],tempc[8],tempc[10],.5);
for(i=0;i<7;i+=2){
interp_curve(c[i+1],c[i],c[i+2],.5);
interp_curve(tempc[i+1],tempc[i],tempc[i+2],.5);
}
/* take things out of dB domain into linear amplitude */
for(i=0;i<11;i++)
for(i=0;i<9;i++)
linear_curve(c[i]);
for(i=0;i<11;i++)
for(i=0;i<9;i++)
linear_curve(tempc[i]);
/* Now limit the louder curves.
the idea is this: We don't know what the playback attenuation
will be; 0dB moves every time the user twiddles the volume
will be; 0dB SL moves every time the user twiddles the volume
knob. So that means we have to use a single 'most pessimal' curve
for all masking amplitudes, right? Wrong. The *loudest* sound
can be in (we assume) a range of 0-120dB SL. However, sounds
20dB down will be in a range of 0-100, 40dB down is from 0-80,
can be in (we assume) a range of ...+100dB] SL. However, sounds
20dB down will be in a range ...+80], 40dB down is from ...+60],
etc... */
for(i=10;i>=0;i--){
analysis("Craw",oc*100+i,c[i],EHMER_MAX,0,1);
analysis("Ctemp",oc*100+i,tempc[i],EHMER_MAX,0,1);
for(i=8;i>=0;i--){
for(j=0;j<i;j++)
min_curve(c[i],tempc[j]);
analysis("C",oc*100+i,c[i],EHMER_MAX,0,1);
}
}
......@@ -185,23 +185,23 @@ void _vp_psy_init(vorbis_look_psy *p,vorbis_info_psy *vi,int n,long rate){
/* set up the lookups for a given blocksize and sample rate */
/* Vorbis max sample rate is limited by 26 Bark (54kHz) */
set_curve(vi->ath, p->ath,n,rate);
set_curve(ATH_Bark_dB, p->ath,n,rate);
for(i=0;i<n;i++){
double oc=toOC((i+.5)*rate2/n);
int pre=fromOC(oc-.0625)/rate2*n;
int post=fromOC(oc+.0625)/rate2*n+1;
int post=fromOC(oc+.0625)/rate2*n;
p->pre[i]=(pre<0?0:pre);
p->octave[i]=oc;
p->post[i]=(post<n?post:n-1);
p->post[i]=(post<0?0:post);
}
p->curves=malloc(11*sizeof(double **));
for(i=0;i<11;i++)
p->curves[i]=malloc(11*sizeof(double *));
p->curves[i]=malloc(9*sizeof(double *));
for(i=0;i<11;i++)
for(j=0;j<11;j++)
for(j=0;j<9;j++)
p->curves[i][j]=malloc(EHMER_MAX*sizeof(double));
memcpy(p->curves[0][2],tone_250_40dB_SL,sizeof(double)*EHMER_MAX);
......@@ -234,12 +234,22 @@ void _vp_psy_init(vorbis_look_psy *p,vorbis_info_psy *vi,int n,long rate){
memcpy(p->curves[10][6],tone_4000_80dB_SL,sizeof(double)*EHMER_MAX);
memcpy(p->curves[10][8],tone_8000_100dB_SL,sizeof(double)*EHMER_MAX);
setup_curve(p->curves[0],0,vi->curveatt_250Hz,vi->peakatt_250Hz,
vi->peakpre,vi->peakpost);
setup_curve(p->curves[2],2,vi->curveatt_500Hz,vi->peakatt_500Hz,
vi->peakpre,vi->peakpost);
setup_curve(p->curves[4],4,vi->curveatt_1000Hz,vi->peakatt_1000Hz,
vi->peakpre,vi->peakpost);
setup_curve(p->curves[6],6,vi->curveatt_2000Hz,vi->peakatt_2000Hz,
vi->peakpre,vi->peakpost);
setup_curve(p->curves[8],8,vi->curveatt_4000Hz,vi->peakatt_4000Hz,
vi->peakpre,vi->peakpost);
setup_curve(p->curves[10],10,vi->curveatt_8000Hz,vi->peakatt_8000Hz,
vi->peakpre,vi->peakpost);
for(i=1;i<11;i+=2)
for(j=0;j<11;j+=2)
interp_curve(p->curves[i][j],p->curves[i-1][j],p->curves[i+1][j],.5);
for(i=0;i<11;i++)
setup_curve(p->curves[i],i,-35.,-40.,-60.,-80.,-100.,-105.);
for(j=0;j<9;j+=2)
interp_curve_dB(p->curves[i][j],p->curves[i-1][j],p->curves[i+1][j],.5);
}
......@@ -252,7 +262,7 @@ void _vp_psy_clear(vorbis_look_psy *p){
if(p->post)free(p->post);
if(p->curves){
for(i=0;i<11;i++){
for(j=0;j<11;j++)
for(j=0;j<9;j++)
free(p->curves[i][j]);
free(p->curves[i]);
}
......@@ -295,40 +305,41 @@ static double _eights[EHMER_MAX]={
static double seed_peaks(double *floor,int *len,double **curve,
double amp,double specmax,
int *pre,int *post,
int x,int n,int addp){
int x,int n,double specatt){
int i;
int ix=x*_eights[0];
int prevx=(ix<0?-1:pre[ix]);
int prevx=pre[ix];
int nextx;
double ret=0.;
/* make this attenuation adjustable */
int choice=rint((amp-specmax+100.)/10.)-2;
int choice=rint((amp-specmax+specatt)/10.)-2;
if(choice<0)choice=0;
if(choice>8)choice=8;
amp=fromdB(amp);
for(i=0;i<EHMER_MAX;i++){
ix=x*_eights[i];
nextx=(ix<n?post[ix]:n-1);
if(addp){
double lin=curve[choice][i]*amp;
/* Currently uses a n+n = +3dB additivity;
6dB may be more correct in most cases */
lin*=lin;
floor[prevx]+=lin;
floor[nextx]-=lin;
if(i==EHMER_OFFSET || prevx==x)ret+=lin;
if(nextx==x)ret-=lin;
}else{
if(floor[prevx]<curve[choice][i]*amp){
floor[prevx]=curve[choice][i]*amp;
len[prevx]=nextx;
for(i=0;i<EHMER_MAX;i++){
if(prevx<n){
double lin=curve[choice][i];
ix=x*_eights[i];
nextx=(ix<n?post[ix]:n);
if(lin){
/* Currently uses a n+n = +3dB additivity */
lin*=amp;
lin*=lin;
floor[prevx]+=lin;
if(nextx==prevx){
if(nextx+1<n)floor[nextx+1]-=lin;
}else{
if(nextx<n)floor[nextx]-=lin;
}
if(i==EHMER_OFFSET || prevx==x)ret+=lin;
if(nextx==x)ret-=lin;
}
if(len[prevx]<nextx+1)len[prevx]=nextx;
prevx=nextx;
}
prevx=nextx;
}
return(ret);
}
......@@ -342,64 +353,21 @@ static void add_seeds(double *floor,int n){
}
}
static void arbitrate_peak(double *floor,int *len,int to,
double val,int end,int n){
if(to<end){
if(len[to]==to || val>floor[to]){
/* new maximum */
arbitrate_peak(floor,len,end,floor[to],len[to],n);
floor[to]=val;
len[to]=end;
}else
arbitrate_peak(floor,len,len[to],val,end,n);
}
}
static void max_seeds(double *floor,int *len,int n){
int i;
double acc=0.;
int end=0;
for(i=0;i<n;i++){
if(i==end){
acc=0.;
end=0;
}
if(floor[i]>acc){
/* new maximum */
arbitrate_peak(floor,len,len[i],acc,end,n);
acc=floor[i];
end=len[i];
}else{
/* nope, but see if we're maximum later */
arbitrate_peak(floor,len,end,floor[i],len[i],n);
}
floor[i]=(acc>0?todB(acc):-DYNAMIC_RANGE_dB);
}
}
/* octave/dB SL scale for masking curves, Bark/dB SPL scale for ATH.
Why Bark scale for encoding but not masking? Because masking has a
strong harmonic dependancy */
extern int frameno;
void _vp_tone_tone_mask(vorbis_look_psy *p,double *f, double *flr,
int athp, int addp, int decayp, double *decay){
int athp, int decayp, double *decay){
vorbis_info_psy *vi=p->vi;
long n=p->n,i;
double *acctemp=alloca(n*sizeof(double));
double *work=alloca(n*sizeof(double));
double *workdB=alloca(n*sizeof(double));
/*double *max=alloca(n*sizeof(double));
int *maxlen=alloca(n*sizeof(int));*/
double specmax=-DYNAMIC_RANGE_dB;
double acc=0.;
/*memset(maxlen,0,n*sizeof(int));*/
for(i=0;i<n;i++)work[i]=fabs(f[i]);
/* slight smoothing; remember that 1 + 1 = sqrt(2) :-)
for(i=n-1;i>0;i--)work[i]=hypot(work[i],work[i-1]);*/
/* handle decay */
if(decayp){
double decscale=1.-pow(vi->decay_coeff,n);
......@@ -421,13 +389,6 @@ void _vp_tone_tone_mask(vorbis_look_psy *p,double *f, double *flr,
if(workdB[i]>specmax)specmax=workdB[i];
}
/* subtract the absolute threshhold of hearing curve so the Ehmer
curves can be used on the data directly */
for(i=0;i<n;i++){
/*work[i]-=p->ath[i]; clearly incorrect; ehmer's early data is not
on an ATH relative scale */
/*max[i]= -DYNAMIC_RANGE_dB;*/
}
memset(flr,0,sizeof(double)*n);
/* prime the working vector with peak values */
......@@ -438,26 +399,21 @@ void _vp_tone_tone_mask(vorbis_look_psy *p,double *f, double *flr,
int o=rint(p->octave[i]*2.);
if(o<0)o=0;
if(o>10)o=10;
acc+=seed_peaks(flr,NULL,p->curves[o],workdB[i],
specmax,p->pre,p->post,i,n,1);
/*seed_peaks(max,maxlen,p->curves[o],workdB[i],
specmax,p->pre,p->post,i,n,0);*/
specmax,p->pre,p->post,i,n,vi->max_curve_dB);
}
acctemp[i]=acc;
}
/* now, chase curves down from the peak seeds */
add_seeds(flr,n);
/*max_seeds(max,maxlen,n);*/
analysis("Pwork",frameno,workdB,n,1,0);
analysis("Padd",frameno,flr,n,1,0);
analysis("Pacc",frameno,acctemp,n,1,1);
/* analysis("Pmax",frameno,max,n,1,0);*/
for(i=0;i<n;i++)if(flr[i]<p->ath[i]-vi->master_att)flr[i]=p->ath[i]-vi->master_att;
for(i=0;i<n;i++)flr[i]+=6.;
/* mask off the ATH */
if(athp)
for(i=0;i<n;i++)
if(flr[i]<p->ath[i]+vi->ath_att)
flr[i]=p->ath[i]+vi->ath_att;
}
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment