/********************************************************************
* *
- * THIS FILE IS PART OF THE Ogg Vorbis SOFTWARE CODEC SOURCE CODE. *
+ * THIS FILE IS PART OF THE OggVorbis SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS SOURCE IS GOVERNED BY *
- * THE GNU PUBLIC LICENSE 2, WHICH IS INCLUDED WITH THIS SOURCE. *
- * PLEASE READ THESE TERMS DISTRIBUTING. *
+ * THE GNU LESSER/LIBRARY PUBLIC LICENSE, WHICH IS INCLUDED WITH *
+ * THIS SOURCE. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
- * THE OggSQUISH SOURCE CODE IS (C) COPYRIGHT 1994-2000 *
- * by Monty <monty@xiph.org> and The XIPHOPHORUS Company *
+ * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2000 *
+ * by Monty <monty@xiph.org> and the XIPHOPHORUS Company *
* http://www.xiph.org/ *
* *
********************************************************************
function: psychoacoustics not including preecho
- last mod: $Id: psy.c,v 1.21 2000/06/14 01:38:31 xiphmont Exp $
+ last mod: $Id: psy.c,v 1.40 2001/02/02 02:52:34 xiphmont Exp $
********************************************************************/
#include <math.h>
#include <string.h>
#include "vorbis/codec.h"
+#include "codec_internal.h"
#include "masking.h"
#include "psy.h"
#include "lpc.h"
#include "smallft.h"
#include "scales.h"
+#include "misc.h"
-/* Why Bark scale for encoding but not masking? Because masking has a
- strong harmonic dependancy */
+#define NEGINF -9999.f
+
+/* Why Bark scale for encoding but not masking computation? Because
+ masking has a strong harmonic dependancy */
/* 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));
- free(i);
+ _ogg_free(i);
}
}
-/* Set up decibel threshhold slopes on a Bark frequency scale */
-/* the only bit left on a Bark scale. No reason to change it right now */
-static void set_curve(double *ref,double *c,int n, double crate){
+vorbis_info_psy *_vi_psy_copy(vorbis_info_psy *i){
+ vorbis_info_psy *ret=_ogg_malloc(sizeof(vorbis_info_psy));
+ memcpy(ret,i,sizeof(vorbis_info_psy));
+ return(ret);
+}
+
+/* Set up decibel threshold slopes on a Bark frequency scale */
+/* ATH is the only bit left on a Bark scale. No reason to change it
+ right now */
+static void set_curve(float *ref,float *c,int n, float crate){
int i,j=0;
for(i=0;i<MAX_BARK-1;i++){
int endpos=rint(fromBARK(i+1)*2*n/crate);
- double base=ref[i];
+ float base=ref[i];
if(j<endpos){
- double delta=(ref[i+1]-base)/(endpos-j);
+ float delta=(ref[i+1]-base)/(endpos-j);
for(;j<endpos && j<n;j++){
c[j]=base;
base+=delta;
}
}
-static void min_curve(double *c,
- double *c2){
+static void min_curve(float *c,
+ float *c2){
int i;
for(i=0;i<EHMER_MAX;i++)if(c2[i]<c[i])c[i]=c2[i];
}
-static void max_curve(double *c,
- double *c2){
+static void max_curve(float *c,
+ float *c2){
int i;
for(i=0;i<EHMER_MAX;i++)if(c2[i]>c[i])c[i]=c2[i];
}
-static void attenuate_curve(double *c,double att){
+static void attenuate_curve(float *c,float att){
int i;
for(i=0;i<EHMER_MAX;i++)
c[i]+=att;
}
-static void linear_curve(double *c){
- int i;
- for(i=0;i<EHMER_MAX;i++)
- if(c[i]<=-900.)
- c[i]=0.;
- else
- 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){
+static void interp_curve(float *c,float *c1,float *c2,float del){
int i;
for(i=0;i<EHMER_MAX;i++)
- c[i]=c2[i]*del+c1[i]*(1.-del);
+ c[i]=c2[i]*del+c1[i]*(1.f-del);
}
-static void setup_curve(double **c,
- int oc,
- double *curveatt_dB){
+static void setup_curve(float **c,
+ int band,
+ float *curveatt_dB){
int i,j;
- double tempc[9][EHMER_MAX];
- double ath[EHMER_MAX];
+ float ath[EHMER_MAX];
+ float tempc[P_LEVELS][EHMER_MAX];
+
+ memcpy(c[0]+2,c[4]+2,sizeof(float)*EHMER_MAX);
+ memcpy(c[2]+2,c[4]+2,sizeof(float)*EHMER_MAX);
+
+ /* we add back in the ATH to avoid low level curves falling off to
+ -infinity and unneccessarily cutting off high level curves in the
+ curve limiting (last step). But again, remember... a half-band's
+ settings must be valid over the whole band, and it's better to
+ mask too little than too much, so be pessimal. */
for(i=0;i<EHMER_MAX;i++){
- double bark=toBARK(fromOC(oc*.5+(i-EHMER_OFFSET)*.125));
+ float oc_min=band*.5+(i-EHMER_OFFSET)*.125;
+ float oc_max=band*.5+(i-EHMER_OFFSET+1)*.125;
+ float bark=toBARK(fromOC(oc_min));
int ibark=floor(bark);
- double del=bark-ibark;
+ float del=bark-ibark;
+ float ath_min,ath_max;
+
if(ibark<26)
- ath[i]=ATH_Bark_dB[ibark]*(1.-del)+ATH_Bark_dB[ibark+1]*del;
+ ath_min=ATH_Bark_dB[ibark]*(1.f-del)+ATH_Bark_dB[ibark+1]*del;
else
- ath[i]=200;
- }
+ ath_min=ATH_Bark_dB[25];
- memcpy(c[0],c[2],sizeof(double)*EHMER_MAX);
+ bark=toBARK(fromOC(oc_max));
+ ibark=floor(bark);
+ del=bark-ibark;
- /* the temp curves are a bit roundabout, but this is only in
- init. */
- for(i=0;i<5;i++){
- memcpy(tempc[i*2],c[i*2],sizeof(double)*EHMER_MAX);
- attenuate_curve(tempc[i*2],curveatt_dB[i]+(i+1)*20);
- max_curve(tempc[i*2],ath);
- attenuate_curve(tempc[i*2],-(i+1)*20);
- }
+ if(ibark<26)
+ ath_max=ATH_Bark_dB[ibark]*(1.f-del)+ATH_Bark_dB[ibark+1]*del;
+ else
+ ath_max=ATH_Bark_dB[25];
- /* normalize them so the driving amplitude is 0dB */
- for(i=0;i<5;i++){
- attenuate_curve(c[i*2],curveatt_dB[i]);
+ ath[i]=min(ath_min,ath_max);
}
/* The c array is comes in as dB curves at 20 40 60 80 100 dB.
interpolate intermediate dB curves */
- 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);
+ for(i=1;i<P_LEVELS;i+=2){
+ interp_curve(c[i]+2,c[i-1]+2,c[i+1]+2,.5);
+ }
+
+ /* normalize curves so the driving amplitude is 0dB */
+ /* make temp curves with the ATH overlayed */
+ for(i=0;i<P_LEVELS;i++){
+ attenuate_curve(c[i]+2,curveatt_dB[i]);
+ memcpy(tempc[i],ath,EHMER_MAX*sizeof(float));
+ attenuate_curve(tempc[i],-i*10.f);
+ max_curve(tempc[i],c[i]+2);
}
- /* take things out of dB domain into linear amplitude */
- for(i=0;i<9;i++)
- linear_curve(c[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
20dB down will be in a range ...+80], 40dB down is from ...+60],
etc... */
- for(i=8;i>=0;i--){
- for(j=0;j<i;j++)
- min_curve(c[i],tempc[j]);
+ for(j=1;j<P_LEVELS;j++){
+ min_curve(tempc[j],tempc[j-1]);
+ min_curve(c[j]+2,tempc[j]);
}
-}
+ /* add fenceposts */
+ for(j=0;j<P_LEVELS;j++){
+
+ for(i=0;i<EHMER_MAX;i++)
+ if(c[j][i+2]>-200.f)break;
+ c[j][0]=i;
+
+ for(i=EHMER_MAX-1;i>=0;i--)
+ if(c[j][i+2]>-200.f)
+ break;
+ c[j][1]=i;
+
+ }
+}
void _vp_psy_init(vorbis_look_psy *p,vorbis_info_psy *vi,int n,long rate){
long i,j;
- double rate2=rate/2.;
+ long maxoc;
memset(p,0,sizeof(vorbis_look_psy));
- p->ath=malloc(n*sizeof(double));
- p->octave=malloc(n*sizeof(int));
+
+
+ p->eighth_octave_lines=vi->eighth_octave_lines;
+ p->shiftoc=rint(log(vi->eighth_octave_lines*8)/log(2))-1;
+
+ p->firstoc=toOC(.25f*rate/n)*(1<<(p->shiftoc+1))-vi->eighth_octave_lines;
+ maxoc=toOC((n*.5f-.25f)*rate/n)*(1<<(p->shiftoc+1))+.5f;
+ p->total_octave_lines=maxoc-p->firstoc+1;
+
+ p->ath=_ogg_malloc(n*sizeof(float));
+ p->octave=_ogg_malloc(n*sizeof(int));
+ p->bark=_ogg_malloc(n*sizeof(float));
p->vi=vi;
p->n=n;
/* set up the lookups for a given blocksize and sample rate */
- /* Vorbis max sample rate is limited by 26 Bark (54kHz) */
+ /* Vorbis max sample rate is currently limited by 26 Bark (54kHz) */
set_curve(ATH_Bark_dB, p->ath,n,rate);
for(i=0;i<n;i++)
- p->ath[i]=fromdB(p->ath[i]+vi->ath_att);
+ p->bark[i]=toBARK(rate/(2*n)*i);
- for(i=0;i<n;i++){
- int oc=rint(toOC((i+.5)*rate2/n)*2.);
- if(oc<0)oc=0;
- if(oc>12)oc=12;
- p->octave[i]=oc;
- }
-
- p->tonecurves=malloc(13*sizeof(double **));
- p->noisecurves=malloc(13*sizeof(double **));
- p->peakatt=malloc(7*sizeof(double *));
- for(i=0;i<13;i++){
- p->tonecurves[i]=malloc(9*sizeof(double *));
- p->noisecurves[i]=malloc(9*sizeof(double *));
+ for(i=0;i<n;i++)
+ p->octave[i]=toOC((i*.5f+.25f)*rate/n)*(1<<(p->shiftoc+1))+.5f;
+
+ p->tonecurves=_ogg_malloc(P_BANDS*sizeof(float **));
+ p->noisemedian=_ogg_malloc(n*sizeof(float));
+ p->noiseoffset=_ogg_malloc(n*sizeof(float));
+ p->peakatt=_ogg_malloc(P_BANDS*sizeof(float *));
+ for(i=0;i<P_BANDS;i++){
+ p->tonecurves[i]=_ogg_malloc(P_LEVELS*sizeof(float *));
+ p->peakatt[i]=_ogg_malloc(P_LEVELS*sizeof(float));
}
- for(i=0;i<7;i++)
- p->peakatt[i]=malloc(5*sizeof(double));
- for(i=0;i<13;i++)
- for(j=0;j<9;j++){
- p->tonecurves[i][j]=malloc(EHMER_MAX*sizeof(double));
- p->noisecurves[i][j]=malloc(EHMER_MAX*sizeof(double));
+ for(i=0;i<P_BANDS;i++)
+ for(j=0;j<P_LEVELS;j++){
+ p->tonecurves[i][j]=_ogg_malloc((EHMER_MAX+2)*sizeof(float));
}
/* OK, yeah, this was a silly way to do it */
- memcpy(p->tonecurves[0][2],tone_125_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[0][4],tone_125_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[0][6],tone_125_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[0][8],tone_125_100dB_SL,sizeof(double)*EHMER_MAX);
-
- memcpy(p->tonecurves[2][2],tone_250_40dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[2][4],tone_250_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[2][6],tone_250_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[2][8],tone_250_80dB_SL,sizeof(double)*EHMER_MAX);
-
- memcpy(p->tonecurves[4][2],tone_500_40dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[4][4],tone_500_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[4][6],tone_500_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[4][8],tone_500_100dB_SL,sizeof(double)*EHMER_MAX);
-
- memcpy(p->tonecurves[6][2],tone_1000_40dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[6][4],tone_1000_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[6][6],tone_1000_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[6][8],tone_1000_100dB_SL,sizeof(double)*EHMER_MAX);
-
- memcpy(p->tonecurves[8][2],tone_2000_40dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[8][4],tone_2000_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[8][6],tone_2000_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[8][8],tone_2000_100dB_SL,sizeof(double)*EHMER_MAX);
-
- memcpy(p->tonecurves[10][2],tone_4000_40dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[10][4],tone_4000_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[10][6],tone_4000_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[10][8],tone_4000_100dB_SL,sizeof(double)*EHMER_MAX);
-
- memcpy(p->tonecurves[12][2],tone_4000_40dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[12][4],tone_4000_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[12][6],tone_8000_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->tonecurves[12][8],tone_8000_100dB_SL,sizeof(double)*EHMER_MAX);
-
-
- memcpy(p->noisecurves[0][2],noise_500_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[0][4],noise_500_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[0][6],noise_500_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[0][8],noise_500_80dB_SL,sizeof(double)*EHMER_MAX);
-
- memcpy(p->noisecurves[2][2],noise_500_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[2][4],noise_500_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[2][6],noise_500_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[2][8],noise_500_80dB_SL,sizeof(double)*EHMER_MAX);
-
- memcpy(p->noisecurves[4][2],noise_500_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[4][4],noise_500_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[4][6],noise_500_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[4][8],noise_500_80dB_SL,sizeof(double)*EHMER_MAX);
-
- memcpy(p->noisecurves[6][2],noise_1000_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[6][4],noise_1000_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[6][6],noise_1000_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[6][8],noise_1000_80dB_SL,sizeof(double)*EHMER_MAX);
-
- memcpy(p->noisecurves[8][2],noise_2000_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[8][4],noise_2000_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[8][6],noise_2000_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[8][8],noise_2000_80dB_SL,sizeof(double)*EHMER_MAX);
-
- memcpy(p->noisecurves[10][2],noise_4000_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[10][4],noise_4000_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[10][6],noise_4000_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[10][8],noise_4000_80dB_SL,sizeof(double)*EHMER_MAX);
-
- memcpy(p->noisecurves[12][2],noise_4000_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[12][4],noise_4000_60dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[12][6],noise_4000_80dB_SL,sizeof(double)*EHMER_MAX);
- memcpy(p->noisecurves[12][8],noise_4000_80dB_SL,sizeof(double)*EHMER_MAX);
-
- setup_curve(p->tonecurves[0],0,vi->toneatt_125Hz);
- setup_curve(p->tonecurves[2],2,vi->toneatt_250Hz);
- setup_curve(p->tonecurves[4],4,vi->toneatt_500Hz);
- setup_curve(p->tonecurves[6],6,vi->toneatt_1000Hz);
- setup_curve(p->tonecurves[8],8,vi->toneatt_2000Hz);
- setup_curve(p->tonecurves[10],10,vi->toneatt_4000Hz);
- setup_curve(p->tonecurves[12],12,vi->toneatt_8000Hz);
-
- setup_curve(p->noisecurves[0],0,vi->noiseatt_125Hz);
- setup_curve(p->noisecurves[2],2,vi->noiseatt_250Hz);
- setup_curve(p->noisecurves[4],4,vi->noiseatt_500Hz);
- setup_curve(p->noisecurves[6],6,vi->noiseatt_1000Hz);
- setup_curve(p->noisecurves[8],8,vi->noiseatt_2000Hz);
- setup_curve(p->noisecurves[10],10,vi->noiseatt_4000Hz);
- setup_curve(p->noisecurves[12],12,vi->noiseatt_8000Hz);
-
- for(i=1;i<13;i+=2)
- for(j=0;j<9;j++){
- interp_curve_dB(p->tonecurves[i][j],
- p->tonecurves[i-1][j],
- p->tonecurves[i+1][j],.5);
- interp_curve_dB(p->noisecurves[i][j],
- p->noisecurves[i-1][j],
- p->noisecurves[i+1][j],.5);
+ memcpy(p->tonecurves[0][4]+2,tone_125_40dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[0][6]+2,tone_125_60dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[0][8]+2,tone_125_80dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[0][10]+2,tone_125_100dB_SL,sizeof(float)*EHMER_MAX);
+
+ memcpy(p->tonecurves[2][4]+2,tone_125_40dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[2][6]+2,tone_125_60dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[2][8]+2,tone_125_80dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[2][10]+2,tone_125_100dB_SL,sizeof(float)*EHMER_MAX);
+
+ memcpy(p->tonecurves[4][4]+2,tone_250_40dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[4][6]+2,tone_250_60dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[4][8]+2,tone_250_80dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[4][10]+2,tone_250_100dB_SL,sizeof(float)*EHMER_MAX);
+
+ memcpy(p->tonecurves[6][4]+2,tone_500_40dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[6][6]+2,tone_500_60dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[6][8]+2,tone_500_80dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[6][10]+2,tone_500_100dB_SL,sizeof(float)*EHMER_MAX);
+
+ memcpy(p->tonecurves[8][4]+2,tone_1000_40dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[8][6]+2,tone_1000_60dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[8][8]+2,tone_1000_80dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[8][10]+2,tone_1000_100dB_SL,sizeof(float)*EHMER_MAX);
+
+ memcpy(p->tonecurves[10][4]+2,tone_2000_40dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[10][6]+2,tone_2000_60dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[10][8]+2,tone_2000_80dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[10][10]+2,tone_2000_100dB_SL,sizeof(float)*EHMER_MAX);
+
+ memcpy(p->tonecurves[12][4]+2,tone_4000_40dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[12][6]+2,tone_4000_60dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[12][8]+2,tone_4000_80dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[12][10]+2,tone_4000_100dB_SL,sizeof(float)*EHMER_MAX);
+
+ memcpy(p->tonecurves[14][4]+2,tone_8000_40dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[14][6]+2,tone_8000_60dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[14][8]+2,tone_8000_80dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[14][10]+2,tone_8000_100dB_SL,sizeof(float)*EHMER_MAX);
+
+ memcpy(p->tonecurves[16][4]+2,tone_8000_40dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[16][6]+2,tone_8000_60dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[16][8]+2,tone_8000_80dB_SL,sizeof(float)*EHMER_MAX);
+ memcpy(p->tonecurves[16][10]+2,tone_8000_100dB_SL,sizeof(float)*EHMER_MAX);
+
+ /* interpolate curves between */
+ for(i=1;i<P_BANDS;i+=2)
+ for(j=4;j<P_LEVELS;j+=2){
+ memcpy(p->tonecurves[i][j]+2,p->tonecurves[i-1][j]+2,EHMER_MAX*sizeof(float));
+ /*interp_curve(p->tonecurves[i][j],
+ p->tonecurves[i-1][j],
+ p->tonecurves[i+1][j],.5);*/
+ min_curve(p->tonecurves[i][j]+2,p->tonecurves[i+1][j]+2);
+ }
+
+ /* set up the final curves */
+ for(i=0;i<P_BANDS;i++)
+ setup_curve(p->tonecurves[i],i,vi->toneatt[i]);
+
+ /* set up attenuation levels */
+ for(i=0;i<P_BANDS;i++)
+ for(j=0;j<P_LEVELS;j++){
+ p->peakatt[i][j]=p->vi->peakatt[i][j];
}
- for(i=0;i<5;i++){
- p->peakatt[0][i]=fromdB(p->vi->peakatt_125Hz[i]);
- p->peakatt[1][i]=fromdB(p->vi->peakatt_250Hz[i]);
- p->peakatt[2][i]=fromdB(p->vi->peakatt_500Hz[i]);
- p->peakatt[3][i]=fromdB(p->vi->peakatt_1000Hz[i]);
- p->peakatt[4][i]=fromdB(p->vi->peakatt_2000Hz[i]);
- p->peakatt[5][i]=fromdB(p->vi->peakatt_4000Hz[i]);
- p->peakatt[6][i]=fromdB(p->vi->peakatt_8000Hz[i]);
+
+ /* set up rolling noise median */
+ for(i=0;i<n;i++){
+ float halfoc=toOC((i+.5)*rate/(2.*n))*2.+2.;
+ int inthalfoc;
+ float del;
+
+ if(halfoc<0)halfoc=0;
+ if(halfoc>=P_BANDS-1)halfoc=P_BANDS-1;
+ inthalfoc=(int)halfoc;
+ del=halfoc-inthalfoc;
+
+ p->noisemedian[i]=
+ p->vi->noisemedian[inthalfoc*2]*(1.-del) +
+ p->vi->noisemedian[inthalfoc*2+2]*del;
+ p->noiseoffset[i]=
+ p->vi->noisemedian[inthalfoc*2+1]*(1.-del) +
+ p->vi->noisemedian[inthalfoc*2+3]*del;
}
+ /*_analysis_output("mediancurve",0,p->noisemedian,n,0,0);*/
}
void _vp_psy_clear(vorbis_look_psy *p){
int i,j;
if(p){
- if(p->ath)free(p->ath);
- if(p->octave)free(p->octave);
- if(p->noisecurves){
- for(i=0;i<13;i++){
- for(j=0;j<9;j++){
- free(p->tonecurves[i][j]);
- free(p->noisecurves[i][j]);
+ if(p->ath)_ogg_free(p->ath);
+ if(p->octave)_ogg_free(p->octave);
+ if(p->bark)_ogg_free(p->bark);
+ if(p->tonecurves){
+ for(i=0;i<P_BANDS;i++){
+ for(j=0;j<P_LEVELS;j++){
+ _ogg_free(p->tonecurves[i][j]);
}
- free(p->noisecurves[i]);
- free(p->tonecurves[i]);
+ _ogg_free(p->tonecurves[i]);
+ _ogg_free(p->peakatt[i]);
}
- for(i=0;i<7;i++)
- free(p->peakatt[i]);
- free(p->tonecurves);
- free(p->noisecurves);
- free(p->peakatt);
+ _ogg_free(p->tonecurves);
+ _ogg_free(p->noisemedian);
+ _ogg_free(p->noiseoffset);
+ _ogg_free(p->peakatt);
}
memset(p,0,sizeof(vorbis_look_psy));
}
}
-static void compute_decay(vorbis_look_psy *p,double *f, double *decay, int n){
- /* handle decay */
+/* octave/(8*eighth_octave_lines) x scale and dB y scale */
+static void seed_curve(float *seed,
+ float **curves,
+ float amp,
+ int oc,int n,int linesper,float dBoffset){
int i;
- double decscale=1.-pow(p->vi->decay_coeff,n);
- double attscale=1.-pow(p->vi->attack_coeff,n);
- for(i=0;i<n;i++){
- double del=f[i]-decay[i];
- if(del>0)
- /* add energy */
- decay[i]+=del*attscale;
- else
- /* remove energy */
- decay[i]+=del*decscale;
- if(decay[i]>f[i])f[i]=decay[i];
+ long seedptr;
+ float *posts,*curve;
+
+ int choice=(int)((amp+dBoffset)*.1f);
+ choice=max(choice,0);
+ choice=min(choice,P_LEVELS-1);
+ posts=curves[choice];
+ curve=posts+2;
+ seedptr=oc+(posts[0]-16)*linesper-(linesper>>1);
+
+ for(i=posts[0];i<posts[1];i++){
+ if(seedptr>0){
+ float lin=amp+curve[i];
+ if(seed[seedptr]<lin)seed[seedptr]=lin;
+ }
+ seedptr+=linesper;
+ if(seedptr>=n)break;
}
}
-static double _eights[EHMER_MAX+1]={
- .2394004745,.2610680627,.2846967347,.3104639837,
- .3385633673,.3692059617,.4026219471,.4390623367,
- .4788008625,.5221360312,.5693933667,.6209278553,
- .6771266124,.7384117901,.8052437489,.8781245150,
- .9576015522,1.0442718740,1.1387865279,1.2418554865,
- 1.3542529803,1.4768233137,1.6104872070,1.7562487129,
- 1.9152027587,2.0885433709,2.2775726445,2.4837105245,
- 2.7085054716,2.9536460940,3.2209738324,3.5124967917,
- 3.8304048259,4.1770859876,4.5551444666,4.9674201522,
- 5.4170099651,5.9072911215,6.4419465017,7.0249923150,
- 7.6608082685,8.3541704668,9.1102872884,9.9348385106,
- 10.8340179740,11.8145801099,12.8838906772,14.0499820932,
- 15.3216137706,16.7083379167,18.2205712869,19.8696734335,
- 21.6680320357,23.6291559533,25.7677767018,28.0999591127,
- 30.6432220084};
-
-static void seed_curve(double *flr,
- double **curves,
- double amp,double specmax,
- int x,int n,double specatt){
- int i;
- int prevx=x*_eights[0]+.5;
- int nextx;
+static void seed_peak(float *seed,
+ float *att,
+ float amp,
+ int oc,
+ int linesper,
+ float dBoffset){
+ long seedptr;
- /* make this attenuation adjustable */
- int choice=(int)((todB(amp)-specmax+specatt)/10.-1.5);
+ int choice=(int)((amp+dBoffset)*.1f);
choice=max(choice,0);
- choice=min(choice,8);
+ choice=min(choice,P_LEVELS-1);
+ seedptr=oc-(linesper>>1);
- for(i=0;i<EHMER_MAX;i++){
- if(prevx<n){
- double lin=curves[choice][i];
- nextx=x*_eights[i+1]+.5;
- nextx=(nextx<n?nextx:n);
- if(lin){
- lin*=amp;
- if(prevx<0){
- if(nextx>=0){
- flr[0]=max(flr[0],lin);
- }
- }else{
- flr[prevx]=max(flr[prevx],lin);
- }
- }
- prevx=nextx;
- }
- }
-}
+ amp+=att[choice];
+ if(seed[seedptr]<amp)seed[seedptr]=amp;
-static void seed_peak(double *flr,
- double *att,
- double amp,double specmax,
- int x,int n,double specatt){
- int prevx=x*_eights[16];
- int nextx=x*_eights[17];
-
- /* make this attenuation adjustable */
- int choice=rint((todB(amp)-specmax+specatt)/20.)-1;
- if(choice<0)choice=0;
- if(choice>4)choice=4;
-
- if(prevx<n){
- double lin=att[choice];
- if(lin){
- lin*=amp;
- if(prevx<0){
- if(nextx>=0){
- if(flr[0]<lin)flr[0]=lin;
- }
- }else{
- if(flr[prevx]<lin)flr[prevx]=lin;
- }
- }
- }
}
-static void seed_generic(vorbis_look_psy *p,
- double ***curves,
- double *f,
- double *flr,
- double specmax){
+static void seed_loop(vorbis_look_psy *p,
+ float ***curves,
+ float **att,
+ float *f,
+ float *flr,
+ float *minseed,
+ float *maxseed,
+ float specmax){
vorbis_info_psy *vi=p->vi;
long n=p->n,i;
-
+ float dBoffset=vi->max_curve_dB-specmax;
+
/* prime the working vector with peak values */
- /* Use the 125 Hz curve up to 125 Hz and 8kHz curve after 8kHz. */
- for(i=0;i<n;i++)
- if(f[i]>flr[i])
- seed_curve(flr,curves[p->octave[i]],f[i],
- specmax,i,n,vi->max_curve_dB);
+
+ for(i=0;i<n;i++){
+ float max=f[i];
+ long oc=p->octave[i];
+ while(i+1<n && p->octave[i+1]==oc){
+ i++;
+ if(f[i]>max)max=f[i];
+ }
+
+ if(max>flr[i]){
+ oc=oc>>p->shiftoc;
+ if(oc>=P_BANDS)oc=P_BANDS-1;
+ if(oc<0)oc=0;
+ if(vi->tonemaskp)
+ seed_curve(minseed,
+ curves[oc],
+ max,
+ p->octave[i]-p->firstoc,
+ p->total_octave_lines,
+ p->eighth_octave_lines,
+ dBoffset);
+ if(vi->peakattp)
+ seed_peak(maxseed,
+ att[oc],
+ max,
+ p->octave[i]-p->firstoc,
+ p->eighth_octave_lines,
+ dBoffset);
+ }
+ }
}
-static void seed_att(vorbis_look_psy *p,
- double *f,
- double *flr,
- double specmax){
- vorbis_info_psy *vi=p->vi;
+static void bound_loop(vorbis_look_psy *p,
+ float *f,
+ float *seeds,
+ float *flr,
+ float att){
long n=p->n,i;
-
- for(i=0;i<n;i++)
- if(f[i]>flr[i])
- seed_peak(flr,p->peakatt[(p->octave[i]+1)>>1],f[i],
- specmax,i,n,vi->max_curve_dB);
+
+ long off=(p->eighth_octave_lines>>1)+p->firstoc;
+ long *ocp=p->octave;
+
+ for(i=0;i<n;i++){
+ long oc=ocp[i]-off;
+ float v=f[i]+att;
+ if(seeds[oc]<v)seeds[oc]=v;
+ }
}
-/* bleaugh, this is more complicated than it needs to be */
-static void max_seeds(vorbis_look_psy *p,double *flr){
- long n=p->n,i,j;
- long *posstack=alloca(n*sizeof(long));
- double *ampstack=alloca(n*sizeof(double));
- long stack=0;
+static void seed_chase(float *seeds, int linesper, long n){
+ long *posstack=alloca(n*sizeof(long));
+ float *ampstack=alloca(n*sizeof(float));
+ long stack=0;
+ long pos=0;
+ long i;
for(i=0;i<n;i++){
if(stack<2){
posstack[stack]=i;
- ampstack[stack++]=flr[i];
+ ampstack[stack++]=seeds[i];
}else{
while(1){
- if(
flr[i]<ampstack[stack-1]){
+ if(
seeds[i]<ampstack[stack-1]){
posstack[stack]=i;
- ampstack[stack++]=flr[i];
+ ampstack[stack++]=seeds[i];
break;
}else{
- if(i<posstack[stack-1]*1.0905077080){
- if(stack>1 && ampstack[stack-1]<ampstack[stack-2] &&
- i<posstack[stack-2]*1.0905077080){
+ if(i<posstack[stack-1]+linesper){
+ if(stack>1 && ampstack[stack-1]<=ampstack[stack-2] &&
+ i<posstack[stack-2]+linesper){
/* we completely overlap, making stack-1 irrelevant. pop it */
stack--;
continue;
}
}
posstack[stack]=i;
- ampstack[stack++]=flr[i];
+ ampstack[stack++]=seeds[i];
break;
}
/* the stack now contains only the positions that are relevant. Scan
'em straight through */
- {
- long pos=0;
- for(i=0;i<stack;i++){
- long endpos;
- if(i<stack-1 && ampstack[i+1]>ampstack[i]){
- endpos=posstack[i+1];
- }else{
- endpos=posstack[i]*1.0905077080+1; /* +1 is important, else bin 0 is
- discarded in short frames */
- }
- if(endpos>n)endpos=n;
- for(j=pos;j<endpos;j++)flr[j]=ampstack[i];
- pos=endpos;
- }
- }
+ for(i=0;i<stack;i++){
+ long endpos;
+ if(i<stack-1 && ampstack[i+1]>ampstack[i]){
+ endpos=posstack[i+1];
+ }else{
+ endpos=posstack[i]+linesper+1; /* +1 is important, else bin 0 is
+ discarded in short frames */
+ }
+ if(endpos>n)endpos=n;
+ for(;pos<endpos;pos++)
+ seeds[pos]=ampstack[i];
+ }
+
/* there. Linear time. I now remember this was on a problem set I
had in Grad Skool... I didn't solve it at the time ;-) */
+
+}
+
+/* bleaugh, this is more complicated than it needs to be */
+static void max_seeds(vorbis_look_psy *p,float *minseed,float *maxseed,
+ float *flr){
+ long n=p->total_octave_lines;
+ int linesper=p->eighth_octave_lines;
+ long linpos=0;
+ long pos;
+
+ seed_chase(minseed,linesper,n); /* for masking */
+ seed_chase(maxseed,linesper,n); /* for peak att */
+
+ pos=p->octave[0]-p->firstoc-(linesper>>1);
+ while(linpos+1<p->n){
+ float min=minseed[pos];
+ float max=maxseed[pos];
+ long end=((p->octave[linpos]+p->octave[linpos+1])>>1)-p->firstoc;
+ while(pos+1<=end){
+ pos++;
+ if((minseed[pos]>NEGINF && minseed[pos]<min) || min==NEGINF)
+ min=minseed[pos];
+ if(maxseed[pos]>max)max=maxseed[pos];
+ }
+ if(max<min)max=min;
+
+ /* seed scale is log. Floor is linear. Map back to it */
+ end=pos+p->firstoc;
+ for(;linpos<p->n && p->octave[linpos]<=end;linpos++)
+ if(flr[linpos]<max)flr[linpos]=max;
+ }
+
+ {
+ float min=minseed[p->total_octave_lines-1];
+ float max=maxseed[p->total_octave_lines-1];
+ if(max<min)max=min;
+ for(;linpos<p->n;linpos++)
+ if(flr[linpos]<max)flr[linpos]=max;
+ }
+
}
-#define noiseBIAS 5
-static void quarter_octave_noise(vorbis_look_psy *p,double *f,double *noise){
- long i,n=p->n;
- long lo=0,hi=0;
- double acc=0.;
+/* quarter-dB bins */
+#define BIN(x) ((int)((x)*negFour))
+#define BINdB(x) ((x)*negQuarter)
+#define BINCOUNT (200*4)
+#define LASTBIN (BINCOUNT-1)
+
+static void bark_noise_median(long n,float *b,float *f,float *noise,
+ float lowidth,float hiwidth,
+ int lomin,int himin,
+ float *thresh,float *off){
+ long i=0,lo=0,hi=0;
+ float bi,threshi;
+ long median=LASTBIN;
+ float negFour = -4.0f;
+ float negQuarter = -0.25f;
+
+ /* these are really integral values, but we store them in floats to
+ avoid excessive float/int conversions, which GCC and MSVC are
+ farily poor at optimizing. */
+
+ float radix[BINCOUNT];
+ float countabove=0;
+ float countbelow=0;
+
+ memset(radix,0,sizeof(radix));
for(i=0;i<n;i++){
- /* not exactly correct, (the center frequency should be centered
- on a *log* scale), but not worth quibbling */
- long newhi=i*_eights[18]+noiseBIAS;
- long newlo=i*_eights[15]-noiseBIAS;
- if(newhi>n)newhi=n;
-
- for(;lo<newlo;lo++)
- acc-=todB(f[lo]); /* yeah, this ain't RMS */
- for(;hi<newhi;hi++)
- acc+=todB(f[hi]);
- noise[i]=fromdB(acc/(hi-lo));
+ /* find new lo/hi */
+ bi=b[i]+hiwidth;
+ for(;hi<n && (hi<i+himin || b[hi]<=bi);hi++){
+ int bin=BIN(f[hi]);
+ if(bin>LASTBIN)bin=LASTBIN;
+ if(bin<0)bin=0;
+ radix[bin]++;
+ if(bin<median)
+ countabove++;
+ else
+ countbelow++;
+ }
+ bi=b[i]-lowidth;
+ for(;lo<i && lo+lomin<i && b[lo]<=bi;lo++){
+ int bin=BIN(f[lo]);
+ if(bin>LASTBIN)bin=LASTBIN;
+ if(bin<0)bin=0;
+ radix[bin]--;
+ if(bin<median)
+ countabove--;
+ else
+ countbelow--;
+ }
+
+ /* move the median if needed */
+ if(countabove+countbelow){
+ threshi = thresh[i]*(countabove+countbelow);
+
+ while(threshi>countbelow && median>0){
+ median--;
+ countabove-=radix[median];
+ countbelow+=radix[median];
+ }
+
+ while(threshi<(countbelow-radix[median]) &&
+ median<LASTBIN){
+ countabove+=radix[median];
+ countbelow-=radix[median];
+ median++;
+ }
+ }
+ noise[i]=BINdB(median)+off[i];
}
-}
-/* stability doesn't matter */
-static int comp(const void *a,const void *b){
- if(fabs(**(double **)a)<fabs(**(double **)b))
- return(1);
- else
- return(-1);
}
-/* move ath and absolute masking to 'apply_floor' to avoid confusion
- with noise fitting and a floor that warbles due to bad LPC fit */
-static int frameno=-1;
-void _vp_compute_mask(vorbis_look_psy *p,double *f,
- double *flr,
- double *decay){
- double *work=alloca(sizeof(double)*p->n);
- double *work2=alloca(sizeof(double)*p->n);
+float _vp_compute_mask(vorbis_look_psy *p,
+ float *fft,
+ float *mdct,
+ float *flr,
+ float *decay,
+ float specmax){
int i,n=p->n;
- double specmax=0.;
+ float localmax=NEGINF;
+ static int seq=0;
- frameno++;
- memset(flr,0,n*sizeof(double));
+ float *minseed=alloca(sizeof(float)*p->total_octave_lines);
+ float *maxseed=alloca(sizeof(float)*p->total_octave_lines);
+ for(i=0;i<p->total_octave_lines;i++)minseed[i]=maxseed[i]=NEGINF;
- for(i=0;i<n;i++)work[i]=fabs(f[i]);
-
- /* find the highest peak so we know the limits */
+ /* go to dB scale. Also find the highest peak so we know the limits */
for(i=0;i<n;i++){
- if(work[i]>specmax)specmax=work[i];
+ fft[i]=todB(fft[i]);
+ if(fft[i]>localmax)localmax=fft[i];
}
- specmax=todB(specmax);
-
- /* don't use the smoothed data for noise */
- if(p->vi->noisemaskp){
- quarter_octave_noise(p,f,work2);
- seed_generic(p,p->noisecurves,work2,flr,specmax);
+ if(specmax<localmax)specmax=localmax;
+
+
+ for(i=0;i<n;i++){
+ mdct[i]=todB(mdct[i]);
}
-
- /* ... or peak att */
- if(p->vi->peakattp)
- seed_att(p,work,flr,specmax);
-
- if(p->vi->smoothp){
- /* compute power^.5 of three neighboring bins to smooth for peaks
- that get split twixt bins/peaks that nail the bin. This evens
- out treatment as we're not doing additive masking any longer. */
- double acc=work[0]*work[0]+work[1]*work[1];
- double prev=work[0];
-
- work[0]=sqrt(acc);
- for(i=1;i<n-1;i++){
- double this=work[i];
- acc+=work[i+1]*work[i+1];
- work[i]=sqrt(acc);
- acc-=prev*prev;
- prev=this;
- }
- work[n-1]=sqrt(acc);
+
+ _analysis_output("mdct",seq,mdct,n,0,0);
+ _analysis_output("fft",seq,fft,n,0,0);
+
+ /* noise masking */
+ if(p->vi->noisemaskp){
+ bark_noise_median(n,p->bark,mdct,flr,
+ p->vi->noisewindowlo,
+ p->vi->noisewindowhi,
+ p->vi->noisewindowlomin,
+ p->vi->noisewindowhimin,
+ p->noisemedian,
+ p->noiseoffset);
+ /* suppress any noise curve > specmax+p->vi->noisemaxsupp */
+ for(i=0;i<n;i++)
+ if(flr[i]>specmax+p->vi->noisemaxsupp)
+ flr[i]=specmax+p->vi->noisemaxsupp;
+ _analysis_output("noise",seq,flr,n,0,0);
+ }else{
+ for(i=0;i<n;i++)flr[i]=NEGINF;
}
-
- /* seed the tone masking */
- if(p->vi->tonemaskp){
- if(p->vi->decayp){
-
- memset(work2,0,n*sizeof(double));
- seed_generic(p,p->tonecurves,work,work2,specmax);
-
- /* chase the seeds */
- max_seeds(p,flr);
- max_seeds(p,work2);
-
- /* compute, update and apply decay accumulator */
- compute_decay(p,work2,decay,n);
- for(i=0;i<n;i++)if(flr[i]<work2[i])flr[i]=work2[i];
-
- }else{
-
- seed_generic(p,p->tonecurves,work,flr,specmax);
-
- /* chase the seeds */
- max_seeds(p,flr);
+ /* set the ATH (floating below localmax, not global max by a
+ specified att) */
+ if(p->vi->athp){
+ float att=localmax+p->vi->ath_adjatt;
+ if(att<p->vi->ath_maxatt)att=p->vi->ath_maxatt;
+
+ for(i=0;i<n;i++){
+ float av=p->ath[i]+att;
+ if(av>flr[i])flr[i]=av;
}
- }else{
- max_seeds(p,flr);
}
+
+ _analysis_output("ath",seq,flr,n,0,0);
+
+ /* tone/peak masking */
+
+ /* XXX apply decay to the fft here */
+
+ seed_loop(p,p->tonecurves,p->peakatt,fft,flr,minseed,maxseed,specmax);
+ bound_loop(p,mdct,maxseed,flr,p->vi->bound_att_dB);
+ _analysis_output("minseed",seq,minseed,p->total_octave_lines,0,0);
+ _analysis_output("maxseed",seq,maxseed,p->total_octave_lines,0,0);
+ max_seeds(p,minseed,maxseed,flr);
+ _analysis_output("final",seq,flr,n,0,0);
+
+ /* doing this here is clean, but we need to find a faster way to do
+ it than to just tack it on */
+
+ for(i=0;i<n;i++)if(mdct[i]>=flr[i])break;
+ if(i==n)for(i=0;i<n;i++)flr[i]=NEGINF;
+
+
+ seq++;
+
+ return(specmax);
}
/* this applies the floor and (optionally) tries to preserve noise
energy in low resolution portions of the spectrum */
/* f and flr are *linear* scale, not dB */
-void _vp_apply_floor(vorbis_look_psy *p,double *f, double *flr){
- double *work=alloca(p->n*sizeof(double));
- double thresh=fromdB(p->vi->noisefit_threshdB);
- int i,j,addcount=0;
- thresh*=thresh;
+void _vp_apply_floor(vorbis_look_psy *p,float *f, float *flr){
+ float *work=alloca(p->n*sizeof(float));
+ int j;
/* subtract the floor */
for(j=0;j<p->n;j++){
if(flr[j]<=0)
- work[j]=0.;
+ work[j]=0.f;
else
work[j]=f[j]/flr[j];
}
- /* mask off the ATH. This should be floating below specmax too, but
- for now, 0dB is fixed... */
- if(p->vi->athp)
- for(j=0;j<p->n;j++)
- if(fabs(f[j])<p->ath[j]){
- /* zeroes can cause rounding stability issues */
- if(f[j]>0)
- work[j]=.1;
- else
- if(f[j]<0)
- work[j]=-.1;
- }
-
- /* look at spectral energy levels. Noise is noise; sensation level
- is important */
- if(p->vi->noisefitp){
- double **index=alloca(p->vi->noisefit_subblock*sizeof(double *));
-
- /* we're looking for zero values that we want to reinstate (to
- floor level) in order to raise the SL noise level back closer
- to original. Desired result; the SL of each block being as
- close to (but still less than) the original as possible. Don't
- bother if the net result is a change of less than
- p->vi->noisefit_thresh dB */
- for(i=0;i<p->n;){
- double original_SL=0.;
- double current_SL=0.;
- int z=0;
-
- /* compute current SL */
- for(j=0;j<p->vi->noisefit_subblock && i<p->n;j++,i++){
- double y=(f[i]*f[i]);
- original_SL+=y;
- if(work[i]){
- current_SL+=y;
- }else{
- if(p->vi->athp){
- if(fabs(f[j])>=p->ath[j])index[z++]=f+i;
- }else
- index[z++]=f+i;
- }
- }
+ memcpy(f,work,p->n*sizeof(float));
+}
- /* sort the values below mask; add back the largest first, stop
- when we violate the desired result above (which may be
- immediately) */
- if(z && current_SL*thresh<original_SL){
- qsort(index,z,sizeof(double *),&comp);
-
- for(j=0;j<z;j++){
- int p=index[j]-f;
- double val=flr[p]*flr[p]+current_SL;
-
- if(val<original_SL){
- addcount++;
- if(f[p]>0)
- work[p]=1;
- else
- work[p]=-1;
- current_SL=val;
- }else
- break;
- }
- }
- }
- }
+float _vp_ampmax_decay(float amp,vorbis_dsp_state *vd){
+ vorbis_info *vi=vd->vi;
+ codec_setup_info *ci=vi->codec_setup;
+ int n=ci->blocksizes[vd->W]/2;
+ float secs=(float)n/vi->rate;
- memcpy(f,work,p->n*sizeof(double));
+ amp+=secs*ci->ampmax_att_per_sec;
+ if(amp<-9999)amp=-9999;
+ return(amp);
}
+
projects / platform / upstream / libvorbis.git / blobdiff