/********************************************************************
* *
- * 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 1994-1999 *
- * by 1999 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: random psychoacoustics (not including preecho)
- author: Monty <xiphmont@mit.edu>
- modifications by: Monty
- last modification date: Aug 26 1999
+ function: psychoacoustics not including preecho
+ last mod: $Id: psy.c,v 1.40 2001/02/02 02:52:34 xiphmont Exp $
********************************************************************/
+#include <stdlib.h>
#include <math.h>
#include <string.h>
-#include "stdio.h"
-#include "codec.h"
+#include "vorbis/codec.h"
+#include "codec_internal.h"
+
+#include "masking.h"
#include "psy.h"
+#include "os.h"
#include "lpc.h"
#include "smallft.h"
-#include "xlogmap.h"
+#include "scales.h"
+#include "misc.h"
+
+#define NEGINF -9999.f
-#define NOISEdB -6
+/* 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));
+ _ogg_free(i);
+ }
+}
-#define MASKdB 20
-#define HROLL 60
-#define LROLL 90
-#define MASKBIAS 10
+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);
+}
-#define LNOISE .95
-#define HNOISE 1.01
-#define NOISEBIAS 20
+/* 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);
+ float base=ref[i];
+ if(j<endpos){
+ float delta=(ref[i+1]-base)/(endpos-j);
+ for(;j<endpos && j<n;j++){
+ c[j]=base;
+ base+=delta;
+ }
+ }
+ }
+}
-/* Find the mean log energy of a given 'band'; used to evaluate tones
- against background noise */
+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(float *c,
+ float *c2){
+ int i;
+ for(i=0;i<EHMER_MAX;i++)if(c2[i]>c[i])c[i]=c2[i];
+}
-/* This is faster than a real convolution, gives us roughly the log f
- scale we seek, and gives OK results. So, that means it's a good
- hack */
+static void attenuate_curve(float *c,float att){
+ int i;
+ for(i=0;i<EHMER_MAX;i++)
+ c[i]+=att;
+}
-/* To add: f scale noise attenuation curve */
+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.f-del);
+}
-void _vp_noise_floor(double *f, double *m,int n){
- long lo=0,hi=0;
- double acc=0,div=0;
+static void setup_curve(float **c,
+ int band,
+ float *curveatt_dB){
int i,j;
+ 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++){
+ 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);
+ float del=bark-ibark;
+ float ath_min,ath_max;
+
+ if(ibark<26)
+ ath_min=ATH_Bark_dB[ibark]*(1.f-del)+ATH_Bark_dB[ibark+1]*del;
+ else
+ ath_min=ATH_Bark_dB[25];
- for(i=100;i<n;i++){
- long newlo=i*LNOISE-NOISEBIAS;
- long newhi=i*HNOISE+NOISEBIAS;
- double temp;
-
- if(newhi>n)newhi=n;
- if(newlo<0)newlo=0;
+ bark=toBARK(fromOC(oc_max));
+ ibark=floor(bark);
+ del=bark-ibark;
- for(j=hi;j<newhi;j++){
- acc+=todB(f[j]);
- div++;
- }
- for(j=lo;j<newlo;j++){
- acc-=todB(f[j]);
- div--;
- }
+ 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];
+
+ ath[i]=min(ath_min,ath_max);
+ }
- hi=newhi;
- lo=newlo;
+ /* The c array is comes in as dB curves at 20 40 60 80 100 dB.
+ interpolate intermediate dB curves */
+ 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);
+ }
+
+ /* Now limit the louder curves.
+
+ the idea is this: We don't know what the playback attenuation
+ 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 ...+100dB] SL. However, sounds
+ 20dB down will be in a range ...+80], 40dB down is from ...+60],
+ etc... */
+
+ 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;
- temp=fromdB(acc/div+NOISEdB); /* The NOISEdB constant should be an
- attenuation curve */
- if(m[i]<temp)m[i]=temp;
}
}
-/* figure the masking curve. linear rolloff on a dB scale, adjusted
- by octave */
-void _vp_mask_floor(double *f, double *m,int n){
- double ocSCALE=1./log(2);
- double curmask=-9.e40;
- double curoc=log(MASKBIAS)*ocSCALE;
- long i;
+void _vp_psy_init(vorbis_look_psy *p,vorbis_info_psy *vi,int n,long rate){
+ long i,j;
+ long maxoc;
+ memset(p,0,sizeof(vorbis_look_psy));
+
+
+ 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 currently limited by 26 Bark (54kHz) */
+ set_curve(ATH_Bark_dB, p->ath,n,rate);
+ for(i=0;i<n;i++)
+ p->bark[i]=toBARK(rate/(2*n)*i);
+
+ 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));
+ }
- /* run mask forward then backward */
- for(i=0;i<n;i++){
- double newmask=todB(f[i])-MASKdB;
- double newoc=log(i+MASKBIAS)*ocSCALE;
- double roll=curmask-(newoc-curoc)*HROLL;
- double lroll;
- if(newmask>roll){
- roll=curmask=newmask;
- curoc=newoc;
+ 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));
}
- lroll=fromdB(roll);
- if(m[i]<lroll)m[i]=lroll;
- }
-
- curmask=-9.e40;
- curoc=log(n+MASKBIAS)*ocSCALE;
- for(i=n-1;i>=0;i--){
- double newmask=todB(f[i])-MASKdB;
- double newoc=log(i+MASKBIAS)*ocSCALE;
- double roll=curmask-(curoc-newoc)*LROLL;
- double lroll;
- if(newmask>roll){
- roll=curmask=newmask;
- curoc=newoc;
+
+ /* OK, yeah, this was a silly way to do it */
+ 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];
+ }
+
+ /* 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)_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]);
+ }
+ _ogg_free(p->tonecurves[i]);
+ _ogg_free(p->peakatt[i]);
+ }
+ _ogg_free(p->tonecurves);
+ _ogg_free(p->noisemedian);
+ _ogg_free(p->noiseoffset);
+ _ogg_free(p->peakatt);
}
- lroll=fromdB(roll);
- if(m[i]<lroll)m[i]=lroll;
+ memset(p,0,sizeof(vorbis_look_psy));
}
}
-/* s must be padded at the end with m-1 zeroes */
-static void time_convolve(double *s,double *r,int n,int m){
+/* 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;
-
- for(i=0;i<n;i++){
- int j;
- double acc=0;
+ 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;
+ }
+}
- for(j=0;j<m;j++)
- acc+=s[i+j]*r[m-j-1];
+static void seed_peak(float *seed,
+ float *att,
+ float amp,
+ int oc,
+ int linesper,
+ float dBoffset){
+ long seedptr;
+
+ int choice=(int)((amp+dBoffset)*.1f);
+ choice=max(choice,0);
+ choice=min(choice,P_LEVELS-1);
+ seedptr=oc-(linesper>>1);
+
+ amp+=att[choice];
+ if(seed[seedptr]<amp)seed[seedptr]=amp;
- s[i]=acc;
- }
}
-/*************************************************************************/
-/* Continuous balance analysis/synthesis *********************************/
+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 */
-/* Compute the average continual spectral balance of the given vectors
- (in radians); encode into LPC coefficients */
+ 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];
+ }
-double _vp_balance_compute(double *A, double *B, double *lpc,lpc_lookup *vb){
- /* correlate in time (the response function is small). Log
- frequency scale, small mapping */
+ 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);
+ }
+ }
+}
- int n=vb->n;
- int mapped=vb->ln;
+static void bound_loop(vorbis_look_psy *p,
+ float *f,
+ float *seeds,
+ float *flr,
+ float att){
+ long n=p->n,i;
- /* 256/15 are arbitrary but unimportant to decoding */
- int resp=15;
- int i;
+ long off=(p->eighth_octave_lines>>1)+p->firstoc;
+ long *ocp=p->octave;
- /* This is encode side. Don't think too hard about it */
-
- double workA[mapped+resp];
- double workB[mapped+resp];
- double p[mapped];
- double workC[resp];
+ for(i=0;i<n;i++){
+ long oc=ocp[i]-off;
+ float v=f[i]+att;
+ if(seeds[oc]<v)seeds[oc]=v;
+ }
+}
- memset(workA,0,sizeof(workA));
- memset(workB,0,sizeof(workB));
- memset(workC,0,sizeof(workC));
+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++){
- int j=vb->bscale[i]+(resp>>1);
- double mag_sq=A[i]*A[i]+B[i]*B[i];
- double phi;
-
- if(B[i]==0)
- phi=M_PI/2.;
- else{
- phi=atan(A[i]/B[i]);
- if((A[i]<0 && B[i]>0)||
- (A[i]>0 && B[i]<0)){
- /* rotate II and IV into the first quadrant. III is already there */
- phi+=M_PI/2;
+ if(stack<2){
+ posstack[stack]=i;
+ ampstack[stack++]=seeds[i];
+ }else{
+ while(1){
+ if(seeds[i]<ampstack[stack-1]){
+ posstack[stack]=i;
+ ampstack[stack++]=seeds[i];
+ break;
+ }else{
+ 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++]=seeds[i];
+ break;
+
+ }
}
}
+ }
- workA[j]+=mag_sq*sin(phi);
- workB[j]+=mag_sq*cos(phi);
+ /* the stack now contains only the positions that are relevant. Scan
+ 'em straight through */
+
+ 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 ;-) */
+
+}
- /* prepare convolution vector. Play with a few different shapes */
+/* 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;
+ }
- for(i=0;i<resp;i++){
- workC[i]=sin(M_PI*(i+1)/(float)(resp+1));
- workC[i]*=workC[i];
+ {
+ 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;
}
+
+}
- time_convolve(workA,workC,mapped,resp);
- time_convolve(workB,workC,mapped,resp);
+/* quarter-dB bins */
+#define BIN(x) ((int)((x)*negFour))
+#define BINdB(x) ((x)*negQuarter)
+#define BINCOUNT (200*4)
+#define LASTBIN (BINCOUNT-1)
- {
- double amp1;
+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;
- for(i=0;i<mapped;i++){
- p[i]=atan2(workA[i],workB[i]);
+ memset(radix,0,sizeof(radix));
+
+ for(i=0;i<n;i++){
+ /* 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--;
}
- amp1=sqrt(vorbis_gen_lpc(p,lpc,vb));
+ /* move the median if needed */
+ if(countabove+countbelow){
+ threshi = thresh[i]*(countabove+countbelow);
- return(amp1);
+ 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];
}
}
-/*void _vp_balance_apply(double *A, double *B, double *lpc, double amp,
- lpc_lookup *vb,int divp){
- int i;
- for(i=0;i<vb->n;i++){
- double mag=sqrt(A[i]*A[i]+B[i]*B[i]);
- double del=vorbis_lpc_magnitude(vb->dscale[i],lpc,vb->m)*amp;
- double phi=atan2(A[i],B[i]);
+float _vp_compute_mask(vorbis_look_psy *p,
+ float *fft,
+ float *mdct,
+ float *flr,
+ float *decay,
+ float specmax){
+ int i,n=p->n;
+ float localmax=NEGINF;
+ static int seq=0;
+
+ 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;
+
+ /* go to dB scale. Also find the highest peak so we know the limits */
+ for(i=0;i<n;i++){
+ fft[i]=todB(fft[i]);
+ if(fft[i]>localmax)localmax=fft[i];
+ }
+ if(specmax<localmax)specmax=localmax;
- if(divp)
- phi-=del;
- else
- phi+=del;
- A[i]=mag*sin(phi);
- B[i]=mag*cos(phi);
+ for(i=0;i<n;i++){
+ mdct[i]=todB(mdct[i]);
+ }
+
+ _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;
+ }
+
+ /* 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;
+ }
+ }
+
+ _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,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.f;
+ else
+ work[j]=f[j]/flr[j];
}
- }*/
+
+ memcpy(f,work,p->n*sizeof(float));
+}
+
+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;
+
+ amp+=secs*ci->ampmax_att_per_sec;
+ if(amp<-9999)amp=-9999;
+ return(amp);
+}
+
+
+
projects / platform / upstream / libvorbis.git / blobdiff