自学教程：C++ toBARK函数代码示例

static void floor0_map_lazy_init(vorbis_block      *vb,				 vorbis_info_floor *infoX,				 vorbis_look_floor0 *look){  if(!look->linearmap[vb->W]){    vorbis_dsp_state   *vd=vb->vd;    vorbis_info        *vi=vd->vi;    codec_setup_info   *ci=vi->codec_setup;    vorbis_info_floor0 *info=(vorbis_info_floor0 *)infoX;    int W=vb->W;    int n=ci->blocksizes[W]/2,j;    /* we choose a scaling constant so that:       floor(bark(rate/2-1)*C)=mapped-1     floor(bark(rate/2)*C)=mapped */    float scale=look->ln/toBARK(info->rate/2.f);        /* the mapping from a linear scale to a smaller bark scale is       straightforward.  We do *not* make sure that the linear mapping       does not skip bark-scale bins; the decoder simply skips them and       the encoder may do what it wishes in filling them.  They're       necessary in some mapping combinations to keep the scale spacing       accurate */    look->linearmap[W]=_ogg_malloc((n+1)*sizeof(**look->linearmap));    for(j=0;j<n;j++){      int val=floor( toBARK((info->rate/2.f)/n*j) 		     *scale); /* bark numbers represent band edges */      if(val>=look->ln)val=look->ln-1; /* guard against the approximation */      look->linearmap[W][j]=val;    }    look->linearmap[W][j]=-1;    look->n[W]=n;  }}

VorbisPsy *vorbis_psy_init(int rate, int n){  long i,j,lo=-99,hi=1;  VorbisPsy *p = speex_alloc(sizeof(VorbisPsy));  memset(p,0,sizeof(*p));    p->n = n;  spx_drft_init(&p->lookup, n);  p->bark = speex_alloc(n*sizeof(*p->bark));  p->rate=rate;  p->vi = &example_tuning;  /* BH4 window */  p->window = speex_alloc(sizeof(*p->window)*n);  float a0 = .35875f;  float a1 = .48829f;  float a2 = .14128f;  float a3 = .01168f;  for(i=0;i<n;i++)    p->window[i] = //a0 - a1*cos(2.*M_PI/n*(i+.5)) + a2*cos(4.*M_PI/n*(i+.5)) - a3*cos(6.*M_PI/n*(i+.5));      sin((i+.5)/n * M_PI)*sin((i+.5)/n * M_PI);  /* bark scale lookups */  for(i=0;i<n;i++){    float bark=toBARK(rate/(2*n)*i);         for(;lo+p->vi->noisewindowlomin<i && 	  toBARK(rate/(2*n)*lo)<(bark-p->vi->noisewindowlo);lo++);        for(;hi<=n && (hi<i+p->vi->noisewindowhimin ||	  toBARK(rate/(2*n)*hi)<(bark+p->vi->noisewindowhi));hi++);        p->bark[i]=((lo-1)<<16)+(hi-1);  }  /* set up rolling noise median */  p->noiseoffset=speex_alloc(n*sizeof(*p->noiseoffset));    for(i=0;i<n;i++){    float halfoc=toOC((i+.5)*rate/(2.*n))*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->noiseoffset[i]=      p->vi->noiseoff[inthalfoc]*(1.-del) +       p->vi->noiseoff[inthalfoc+1]*del;      }#if 0  _analysis_output_always("noiseoff0",ls,p->noiseoffset,n,1,0,0);#endif   return p;}

/* there was no great place to put this.... */void _analysis_output_always(char *base,int i,float *v,int n,int bark,int dB,ogg_int64_t off){  int j;  FILE *of;  char buffer[80];  /*  if(i==5870){*/    sprintf(buffer,"%s_%d.m",base,i);    of=fopen(buffer,"w");        if(!of)perror("failed to open data dump file");        for(j=0;j<n;j++){      if(bark){	float b=toBARK((4000.f*j/n)+.25);	fprintf(of,"%f ",b);      }else	if(off!=0)	  fprintf(of,"%f ",(double)(j+off)/8000.);	else	  fprintf(of,"%f ",(double)j);            if(dB){	float val;	if(v[j]==0.)	  val=-140.;	else	  val=todB(v+j);	fprintf(of,"%f/n",val);      }else{	fprintf(of,"%f/n",v[j]);      }    }    fclose(of);    /*  } */}

void analysis(char *base,int i,float *v,int n,int bark,int dB){  if(noisy){    int j;    FILE *of;    char buffer[80];    sprintf(buffer,"%s_%d.m",base,i);    of=fopen(buffer,"w");    for(j=0;j<n;j++){      if(dB && v[j]==0)	  fprintf(of,"/n/n");      else{	if(bark)	  fprintf(of,"%g ",toBARK(22050.f*j/n));	else	  fprintf(of,"%g ",(float)j);      	if(dB){	  fprintf(of,"%g/n",todB(v+j));	}else{	  fprintf(of,"%g/n",v[j]);	}      }    }    fclose(of);  }}

static void _analysis_output(char *base,int i,float *v,int n,int bark,int dB){  int j;  FILE *of;  char buffer[80];  sprintf(buffer,"%s_%d.m",base,i);  of=fopen(buffer,"w");    if(!of)perror("failed to open data dump file");    for(j=0;j<n;j++){    if(bark){      float b=toBARK((4000.f*j/n)+.25);      fprintf(of,"%f ",b);    }else      fprintf(of,"%f ",(double)j);        if(dB){      float val;      if(v[j]==0.)	val=-140.;      else	val=todB(v[j]);      fprintf(of,"%f/n",val);    }else{      fprintf(of,"%f/n",v[j]);    }  }  fclose(of);}

static void floor0_map_lazy_init(vorbis_dsp_state  *vd,				 vorbis_info_floor *infoX,				 vorbis_look_floor0 *look,				 unsigned int W){ if(!look->linearcosmap[W]){  vorbis_info        *vi=vd->vi;  codec_setup_info   *ci=vi->codec_setup;  vorbis_info_floor0 *info=(vorbis_info_floor0 *)infoX;  int n=ci->blocksizes[W]/2,j;  ogg_double_t scale=look->ln/toBARK(info->rate/2.f);  look->linearcosmap[W]=_ogg_malloc((n+1)*sizeof(**look->linearcosmap));  for(j=0;j<n;j++){   int val=floor( toBARK((info->rate/2.f)/n*j)*scale);   if(val>=look->ln)    val=look->ln-1;   look->linearcosmap[W][j]=2.0f*cos(M_PI/(float)look->ln*(float)val);;  }  look->linearcosmap[W][j]=-999.0f;  look->n[W]=n; }}

void _vp_psy_init(vorbis_look_psy *p,vorbis_info_psy *vi,		  vorbis_info_psy_global *gi,int n,long rate){  long i,j,lo=-99,hi=1;  long maxoc;  memset(p,0,sizeof(*p));  p->eighth_octave_lines=gi->eighth_octave_lines;  p->shiftoc=rint(log(gi->eighth_octave_lines*8.f)/log(2.f))-1;  p->firstoc=toOC(.25f*rate*.5/n)*(1<<(p->shiftoc+1))-gi->eighth_octave_lines;  maxoc=toOC((n+.25f)*rate*.5/n)*(1<<(p->shiftoc+1))+.5f;  p->total_octave_lines=maxoc-p->firstoc+1;  p->ath=(float*)_ogg_malloc(n*sizeof(*p->ath));  p->octave=(long*)_ogg_malloc(n*sizeof(*p->octave));  p->bark=(long*)_ogg_malloc(n*sizeof(*p->bark));  p->vi=vi;  p->n=n;  p->rate=rate;  /* AoTuV HF weighting */  p->m_val = 1.;  if(rate < 26000) p->m_val = 0;  else if(rate < 38000) p->m_val = .94;   /* 32kHz */  else if(rate > 46000) p->m_val = 1.275; /* 48kHz */  /* set up the lookups for a given blocksize and sample rate */  for(i=0,j=0;i<MAX_ATH-1;i++){    int endpos=rint(fromOC((i+1)*.125-2.)*2*n/rate);    float base=ATH[i];    if(j<endpos){      float delta=(ATH[i+1]-base)/(endpos-j);      for(;j<endpos && j<n;j++){        p->ath[j]=base+100.;        base+=delta;      }    }  }  for(i=0;i<n;i++){    float bark=toBARK(rate/(2*n)*i);    for(;lo+vi->noisewindowlomin<i &&	  toBARK(rate/(2*n)*lo)<(bark-vi->noisewindowlo);lo++);    for(;hi<=n && (hi<i+vi->noisewindowhimin ||	  toBARK(rate/(2*n)*hi)<(bark+vi->noisewindowhi));hi++);    p->bark[i]=((lo-1)<<16)+(hi-1);  }  for(i=0;i<n;i++)    p->octave[i]=toOC((i+.25f)*.5*rate/n)*(1<<(p->shiftoc+1))+.5f;  p->tonecurves=setup_tone_curves(vi->toneatt,rate*.5/n,n,				  vi->tone_centerboost,vi->tone_decay);  /* set up rolling noise median */  p->noiseoffset=(float**)_ogg_malloc(P_NOISECURVES*sizeof(*p->noiseoffset));  for(i=0;i<P_NOISECURVES;i++)    p->noiseoffset[i]=(float*)_ogg_malloc(n*sizeof(**p->noiseoffset));  for(i=0;i<n;i++){    float halfoc=toOC((i+.5)*rate/(2.*n))*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;    for(j=0;j<P_NOISECURVES;j++)      p->noiseoffset[j][i]=	p->vi->noiseoff[j][inthalfoc]*(1.-del) +	p->vi->noiseoff[j][inthalfoc+1]*del;  }#if 0  {    static int ls=0;    _analysis_output_always("noiseoff0",ls,p->noiseoffset[0],n,1,0,0);    _analysis_output_always("noiseoff1",ls,p->noiseoffset[1],n,1,0,0);    _analysis_output_always("noiseoff2",ls++,p->noiseoffset[2],n,1,0,0);  }#endif}

FilterBank *filterbank_new(int banks, spx_word32_t sampling, int len, int type){   FilterBank *bank;   spx_word32_t df;   spx_word32_t max_mel, mel_interval;   int i;   int id1;   int id2;   df = DIV32(SHL32(sampling,15),MULT16_16(2,len));   max_mel = toBARK(EXTRACT16(sampling/2));   mel_interval = PDIV32(max_mel,banks-1);      bank = (FilterBank*)speex_alloc(sizeof(FilterBank));   bank->nb_banks = banks;   bank->len = len;   bank->bank_left = (int*)speex_alloc(len*sizeof(int));   bank->bank_right = (int*)speex_alloc(len*sizeof(int));   bank->filter_left = (spx_word16_t*)speex_alloc(len*sizeof(spx_word16_t));   bank->filter_right = (spx_word16_t*)speex_alloc(len*sizeof(spx_word16_t));   /* Think I can safely disable normalisation that for fixed-point (and probably float as well) */#ifndef FIXED_POINT   bank->scaling = (float*)speex_alloc(banks*sizeof(float));#endif   for (i=0;i<len;i++)   {      spx_word16_t curr_freq;      spx_word32_t mel;      spx_word16_t val;      curr_freq = EXTRACT16(MULT16_32_P15(i,df));      mel = toBARK(curr_freq);      if (mel > max_mel)         break;#ifdef FIXED_POINT      id1 = DIV32(mel,mel_interval);#else            id1 = (int)(floor(mel/mel_interval));#endif      if (id1>banks-2)      {         id1 = banks-2;         val = Q15_ONE;      } else {         val = DIV32_16(mel - id1*mel_interval,EXTRACT16(PSHR32(mel_interval,15)));      }      id2 = id1+1;      bank->bank_left[i] = id1;      bank->filter_left[i] = SUB16(Q15_ONE,val);      bank->bank_right[i] = id2;      bank->filter_right[i] = val;   }      /* Think I can safely disable normalisation for fixed-point (and probably float as well) */#ifndef FIXED_POINT   for (i=0;i<bank->nb_banks;i++)      bank->scaling[i] = 0;   for (i=0;i<bank->len;i++)   {      int id = bank->bank_left[i];      bank->scaling[id] += bank->filter_left[i];      id = bank->bank_right[i];      bank->scaling[id] += bank->filter_right[i];   }   for (i=0;i<bank->nb_banks;i++)      bank->scaling[i] = Q15_ONE/(bank->scaling[i]);#endif   return bank;}

int main(){  int i;  double rate;  for(i=64;i<32000;i*=2){    rate=48000.f;    fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)/n",            rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2));    rate=44100.f;    fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)/n",            rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2));    rate=32000.f;    fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)/n",            rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2));    rate=22050.f;    fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)/n",            rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2));    rate=16000.f;    fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)/n",            rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2));    rate=11025.f;    fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)/n",            rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2));    rate=8000.f;    fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)/n/n",            rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2));  }  {    float i;    int j;    for(i=0.,j=0;i<28;i+=1,j++){      fprintf(stderr,"(%d) bark=%f %gHz (%d of 128)/n",              j,i,fromBARK(i),(int)(fromBARK(i)/22050.*128.));    }  }  return(0);}

void _vp_psy_init(vorbis_look_psy *p,vorbis_info_psy *vi,		  vorbis_info_psy_global *gi,int n,long rate){  long i,j,k,lo=0,hi=0;  long maxoc;  memset(p,0,sizeof(vorbis_look_psy));  p->eighth_octave_lines=gi->eighth_octave_lines;  p->shiftoc=rint(log(gi->eighth_octave_lines*8)/log(2))-1;  p->firstoc=toOC(.25f*rate/n)*(1<<(p->shiftoc+1))-gi->eighth_octave_lines;  maxoc=toOC((n*.5f-.25f)*rate/n)*(1<<(p->shiftoc+1))+.5f;  p->total_octave_lines=maxoc-p->firstoc+1;  if(vi->ath)    p->ath=_ogg_malloc(n*sizeof(float));  p->octave=_ogg_malloc(n*sizeof(long));  p->bark=_ogg_malloc(n*sizeof(unsigned long));  p->vi=vi;  p->n=n;  p->rate=rate;  /* set up the lookups for a given blocksize and sample rate */  if(vi->ath)    set_curve(vi->ath, p->ath,n,rate);  for(i=0;i<n;i++){    float bark=toBARK(rate/(2*n)*i);     for(;lo+vi->noisewindowlomin<i && 	  toBARK(rate/(2*n)*lo)<(bark-vi->noisewindowlo);lo++);        for(;hi<n && (hi<i+vi->noisewindowhimin ||	  toBARK(rate/(2*n)*hi)<(bark+vi->noisewindowhi));hi++);        p->bark[i]=(hi<<16)+lo;  }  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->noisethresh=_ogg_malloc(n*sizeof(float));  p->noiseoffset=_ogg_malloc(n*sizeof(float));  for(i=0;i<P_BANDS;i++)    p->tonecurves[i]=_ogg_malloc(P_LEVELS*sizeof(float *));    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][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);  /* value limit the tonal masking curves; the peakatt not only     optionally specifies maximum dynamic depth, but also [always]//.........这里部分代码省略.........

static void setup_curve(float **c,			int band,			float *curveatt_dB){  int i,j;  float ath[EHMER_MAX];  float tempc[P_LEVELS][EHMER_MAX];  float *ATH=ATH_Bark_dB_lspconservative; /* just for limiting here */  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[ibark]*(1.f-del)+ATH[ibark+1]*del;    else      ath_min=ATH[25];    bark=toBARK(fromOC(oc_max));    ibark=floor(bark);    del=bark-ibark;    if(ibark<26)      ath_max=ATH[ibark]*(1.f-del)+ATH[ibark+1]*del;    else      ath_max=ATH[25];    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=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_OFFSET;i++)      if(c[j][i+2]>-200.f)break;      c[j][0]=i;    for(i=EHMER_MAX-1;i>EHMER_OFFSET+1;i--)      if(c[j][i+2]>-200.f)	break;    c[j][1]=i;  }}