nb_celp.cpp

KphoneSI (kpsi) is a SIP (Session Initiation Protocol) user agent for Linux, with which you can in

            }
         } else {
            pit_min = st->min_pitch;
            pit_max = st->max_pitch;
         }

         /* Pitch synthesis */
         SUBMODE(ltp_unquant)(exc, pit_min, pit_max, ol_pitch_coef, SUBMODE(ltp_params), 
                              st->subframeSize, &pitch, &pitch_gain[0], bits, stack, st->count_lost, offset, st->last_pitch_gain);
         
         /* If we had lost frames, check energy of last received frame */
         if (st->count_lost && ol_gain < st->last_ol_gain)
         {
            float fact = ol_gain/(st->last_ol_gain+1);
            for (i=0;i<st->subframeSize;i++)
               exc[i]*=fact;
         }

         tmp = fabs(pitch_gain[0]+pitch_gain[1]+pitch_gain[2]);
         tmp = fabs(pitch_gain[1]);
         if (pitch_gain[0]>0)
            tmp += pitch_gain[0];
         else
            tmp -= .5*pitch_gain[0];
         if (pitch_gain[2]>0)
            tmp += pitch_gain[2];
         else
            tmp -= .5*pitch_gain[0];


         pitch_average += tmp;
         if (tmp>best_pitch_gain)
         {
            best_pitch = pitch;
	    best_pitch_gain = tmp;
	    /*            best_pitch_gain = tmp*.9;
	                if (best_pitch_gain>.85)
                        best_pitch_gain=.85;*/
         }
      } else {
         speex_error("No pitch prediction, what's wrong");
      }
      
      /* Unquantize the innovation */
      {
         int q_energy;
         float ener;
         float *innov;
         
         innov = st->innov+sub*st->subframeSize;
         for (i=0;i<st->subframeSize;i++)
            innov[i]=0;

         /* Decode sub-frame gain correction */
         if (SUBMODE(have_subframe_gain)==3)
         {
            q_energy = speex_bits_unpack_unsigned(bits, 3);
            ener = ol_gain*exp(exc_gain_quant_scal3[q_energy]);
         } else if (SUBMODE(have_subframe_gain)==1)
         {
            q_energy = speex_bits_unpack_unsigned(bits, 1);
            ener = ol_gain*exp(exc_gain_quant_scal1[q_energy]);
         } else {
            ener = ol_gain;
         }
                  
         if (SUBMODE(innovation_unquant))
         {
            /*Fixed codebook contribution*/
            SUBMODE(innovation_unquant)(innov, SUBMODE(innovation_params), st->subframeSize, bits, stack);
         } else {
            speex_error("No fixed codebook");
         }

         /* De-normalize innovation and update excitation */
         for (i=0;i<st->subframeSize;i++)
            innov[i]*=ener;

         /*Vocoder mode*/
         if (st->submodeID==1) 
         {
            float g=ol_pitch_coef;

            
            for (i=0;i<st->subframeSize;i++)
               exc[i]=0;
            while (st->voc_offset<st->subframeSize)
            {
               if (st->voc_offset>=0)
                  exc[st->voc_offset]=sqrt(1.0*ol_pitch);
               st->voc_offset+=ol_pitch;
            }
            st->voc_offset -= st->subframeSize;

            g=.5+2*(g-.6);
            if (g<0)
               g=0;
            if (g>1)
               g=1;
            for (i=0;i<st->subframeSize;i++)
            {
               float exci=exc[i];
               exc[i]=.8*g*exc[i]*ol_gain + .6*g*st->voc_m1*ol_gain + .5*g*innov[i] - .5*g*st->voc_m2 + (1-g)*innov[i];
               st->voc_m1 = exci;
               st->voc_m2=innov[i];
               st->voc_mean = .95*st->voc_mean + .05*exc[i];
               exc[i]-=st->voc_mean;
            }
         } else {
            for (i=0;i<st->subframeSize;i++)
               exc[i]+=innov[i];
         }
         /* Decode second codebook (only for some modes) */
         if (SUBMODE(double_codebook))
         {
            char *tmp_stack=stack;
            float *innov2 = PUSH(tmp_stack, st->subframeSize, float);
            for (i=0;i<st->subframeSize;i++)
               innov2[i]=0;
            SUBMODE(innovation_unquant)(innov2, SUBMODE(innovation_params), st->subframeSize, bits, tmp_stack);
            for (i=0;i<st->subframeSize;i++)
               innov2[i]*=ener*(1/2.2);
            for (i=0;i<st->subframeSize;i++)
               exc[i] += innov2[i];
         }

      }

      for (i=0;i<st->subframeSize;i++)
         sp[i]=exc[i];

      /* Signal synthesis */
      if (st->lpc_enh_enabled && SUBMODE(comb_gain)>0)
         comb_filter(exc, sp, st->interp_qlpc, st->lpcSize, st->subframeSize,
                              pitch, pitch_gain, SUBMODE(comb_gain), st->comb_mem);
      if (st->lpc_enh_enabled)
      {
         /* Use enhanced LPC filter */
         filter_mem2(sp, awk2, awk1, sp, st->subframeSize, st->lpcSize, 
                     st->mem_sp+st->lpcSize);
         filter_mem2(sp, awk3, st->interp_qlpc, sp, st->subframeSize, st->lpcSize, 
                     st->mem_sp);
      } else {
         /* Use regular filter */
         for (i=0;i<st->lpcSize;i++)
            st->mem_sp[st->lpcSize+i] = 0;
         iir_mem2(sp, st->interp_qlpc, sp, st->subframeSize, st->lpcSize, 
                     st->mem_sp);
      }
   }
   
   /*Copy output signal*/
   out[0] = st->frame[0] + st->preemph*st->pre_mem;
   for (i=1;i<st->frameSize;i++)
     out[i]=st->frame[i] + st->preemph*out[i-1];
   st->pre_mem=out[st->frameSize-1];


   /* Store the LSPs for interpolation in the next frame */
   for (i=0;i<st->lpcSize;i++)
      st->old_qlsp[i] = st->qlsp[i];

   /* The next frame will not be the first (Duh!) */
   st->first = 0;
   st->count_lost=0;
   st->last_pitch = best_pitch;
   st->last_pitch_gain = .25*pitch_average;
   st->pitch_gain_buf[st->pitch_gain_buf_idx++] = st->last_pitch_gain;
   if (st->pitch_gain_buf_idx > 2) /* rollover */
      st->pitch_gain_buf_idx = 0;

   st->last_ol_gain = ol_gain;

   return 0;
}

int nb_encoder_ctl(void *state, int request, void *ptr)
{
   EncState *st;
   st=(EncState*)state;     
   switch(request)
   {
   case SPEEX_GET_FRAME_SIZE:
      (*(int*)ptr) = st->frameSize;
      break;
   case SPEEX_SET_LOW_MODE:
   case SPEEX_SET_MODE:
      st->submodeSelect = st->submodeID = (*(int*)ptr);
      break;
   case SPEEX_GET_LOW_MODE:
   case SPEEX_GET_MODE:
      (*(int*)ptr) = st->submodeID;
      break;
   case SPEEX_SET_VBR:
      st->vbr_enabled = (*(int*)ptr);
      break;
   case SPEEX_GET_VBR:
      (*(int*)ptr) = st->vbr_enabled;
      break;
   case SPEEX_SET_VAD:
      st->vad_enabled = (*(int*)ptr);
      break;
   case SPEEX_GET_VAD:
      (*(int*)ptr) = st->vad_enabled;
      break;
   case SPEEX_SET_DTX:
      st->dtx_enabled = (*(int*)ptr);
      break;
   case SPEEX_GET_DTX:
      (*(int*)ptr) = st->dtx_enabled;
      break;
   case SPEEX_SET_ABR:
      st->abr_enabled = (*(int*)ptr);
      st->vbr_enabled = 1;
      {
         int i=10, rate, target;
         float vbr_qual;
         target = (*(int*)ptr);
         while (i>=0)
         {
            speex_encoder_ctl(st, SPEEX_SET_QUALITY, &i);
            speex_encoder_ctl(st, SPEEX_GET_BITRATE, &rate);
            if (rate <= target)
               break;
            i--;
         }
         vbr_qual=i;
         if (vbr_qual<0)
            vbr_qual=0;
         speex_encoder_ctl(st, SPEEX_SET_VBR_QUALITY, &vbr_qual);
         st->abr_count=0;
         st->abr_drift=0;
         st->abr_drift2=0;
      }
      
      break;
   case SPEEX_GET_ABR:
      (*(int*)ptr) = st->abr_enabled;
      break;
   case SPEEX_SET_VBR_QUALITY:
      st->vbr_quality = (*(float*)ptr);
      break;
   case SPEEX_GET_VBR_QUALITY:
      (*(float*)ptr) = st->vbr_quality;
      break;
   case SPEEX_SET_QUALITY:
      {
         int quality = (*(int*)ptr);
         if (quality < 0)
            quality = 0;
         if (quality > 10)
            quality = 10;
         st->submodeSelect = st->submodeID = ((SpeexNBMode*)(st->mode->mode))->quality_map[quality];
      }
      break;
   case SPEEX_SET_COMPLEXITY:
      st->complexity = (*(int*)ptr);
      if (st->complexity<1)
         st->complexity=1;
      break;
   case SPEEX_GET_COMPLEXITY:
      (*(int*)ptr) = st->complexity;
      break;
   case SPEEX_SET_BITRATE:
      {
         int i=10, rate, target;
         target = (*(int*)ptr);
         while (i>=0)
         {
            speex_encoder_ctl(st, SPEEX_SET_QUALITY, &i);
            speex_encoder_ctl(st, SPEEX_GET_BITRATE, &rate);
            if (rate <= target)
               break;
            i--;
         }
      }
      break;
   case SPEEX_GET_BITRATE:
      if (st->submodes[st->submodeID])
         (*(int*)ptr) = st->sampling_rate*SUBMODE(bits_per_frame)/st->frameSize;
      else
         (*(int*)ptr) = st->sampling_rate*(NB_SUBMODE_BITS+1)/st->frameSize;
      break;
   case SPEEX_SET_SAMPLING_RATE:
      st->sampling_rate = (*(int*)ptr);
      break;
   case SPEEX_GET_SAMPLING_RATE:
      (*(int*)ptr)=st->sampling_rate;
      break;
   case SPEEX_RESET_STATE:
      {
         int i;
         st->bounded_pitch = 1;
         st->first = 1;
         for (i=0;i<st->lpcSize;i++)
            st->lsp[i]=(M_PI*((float)(i+1)))/(st->lpcSize+1);
         for (i=0;i<st->lpcSize;i++)
            st->mem_sw[i]=st->mem_sw_whole[i]=st->mem_sp[i]=st->mem_exc[i]=0;
         for (i=0;i<st->bufSize;i++)
            st->excBuf[i]=st->swBuf[i]=st->inBuf[i]=st->exc2Buf[i]=0;
      }
      break;
   case SPEEX_GET_LOOKAHEAD:
      (*(int*)ptr)=(st->windowSize-st->frameSize);
      break;
   case SPEEX_GET_PI_GAIN:
      {
         int i;
         float *g = (float*)ptr;
         for (i=0;i<st->nbSubframes;i++)
            g[i]=st->pi_gain[i];
      }
      break;
   case SPEEX_GET_EXC:
      {
         int i;
         float *e = (float*)ptr;
         for (i=0;i<st->frameSize;i++)
            e[i]=st->exc[i];
      }
      break;
   case SPEEX_GET_INNOV:
      {
         int i;
         float *e = (float*)ptr;
         for (i=0;i<st->frameSize;i++)
            e[i]=st->innov[i];
      }
      break;
   case SPEEX_GET_RELATIVE_QUALITY:
      (*(float*)ptr)=st->relative_quality;
      break;
   default:
      speex_warning_int("Unknown nb_ctl request: ", request);
      return -1;
   }
   return 0;
}

int nb_decoder_ctl(void *state, int request, void *ptr)
{
   DecState *st;
   st=(DecState*)state;
   switch(request)
   {
   case SPEEX_GET_LOW_MODE:
   case SPEEX_GET_MODE:
      (*(int*)ptr) = st->submodeID;
      break;
   case SPEEX_SET_ENH:
      st->lpc_enh_enabled = *((int*)ptr);
      break;
   case SPEEX_GET_ENH:
      *((int*)ptr) = st->lpc_enh_enabled;
      break;
   case SPEEX_GET_FRAME_SIZE:
      (*(int*)ptr) = st->frameSize;
      break;
   case SPEEX_GET_BITRATE:
      if (st->submodes[st->submodeID])
         (*(int*)ptr) = st->sampling_rate*SUBMODE(bits_per_frame)/st->frameSize;
      else
         (*(int*)ptr) = st->sampling_rate*(NB_SUBMODE_BITS+1)/st->frameSize;
      break;
   case SPEEX_SET_SAMPLING_RATE:
      st->sampling_rate = (*(int*)ptr);
      break;
   case SPEEX_GET_SAMPLING_RATE:
      (*(int*)ptr)=st->sampling_rate;
      break;
   case SPEEX_SET_HANDLER:
      {
         SpeexCallback *c = (SpeexCallback*)ptr;
         st->speex_callbacks[c->callback_id].func=c->func;
         st->speex_callbacks[c->callback_id].data=c->data;
         st->speex_callbacks[c->callback_id].callback_id=c->callback_id;
      }
      break;
   case SPEEX_SET_USER_HANDLER:
      {
         SpeexCallback *c = (SpeexCallback*)ptr;
         st->user_callback.func=c->func;
         st->user_callback.data=c->data;
         st->user_callback.callback_id=c->callback_id;
      }
      break;
   case SPEEX_RESET_STATE:
      {
         int i;
         for (i=0;i<2*st->lpcSize;i++)
            st->mem_sp[i]=0;
         for (i=0;i<st->bufSize;i++)
            st->excBuf[i]=st->inBuf[i]=0;
      }
      break;
   case SPEEX_GET_PI_GAIN:
      {
         int i;
         float *g = (float*)ptr;
         for (i=0;i<st->nbSubframes;i++)
            g[i]=st->pi_gain[i];
      }
      break;
   case SPEEX_GET_EXC:
      {
         int i;
         float *e = (float*)ptr;
         for (i=0;i<st->frameSize;i++)
            e[i]=st->exc[i];
      }
      break;
   case SPEEX_GET_INNOV:
      {
         int i;
         float *e = (float*)ptr;
         for (i=0;i<st->frameSize;i++)
            e[i]=st->innov[i];
      }
      break;
   case SPEEX_GET_DTX_STATUS:
      *((int*)ptr) = st->dtx_enabled;
      break;
   default:
      speex_warning_int("Unknown nb_ctl request: ", request);
      return -1;
   }
   return 0;
}