How the code calculate the pitch?

[lmingde]

I read the paper and codes and have some trouble about the calculation of pitch?

The code which I think is to calculate the pitch: in lpcnet_enc.c line 526-546

  for (i=0;i<FRAME_SIZE;i++) {
    int j;
    float sum = aligned_in[i];
    for (j=0;j<LPC_ORDER;j++)
      sum += st->lpc[j]*st->pitch_mem[j];
    RNN_MOVE(st->pitch_mem+1, st->pitch_mem, LPC_ORDER-1);
    st->pitch_mem[0] = aligned_in[i];
    st->exc_buf[PITCH_MAX_PERIOD+i] = sum + .7*st->pitch_filt;
    st->pitch_filt = sum;
    //printf("%f\n", st->exc_buf[PITCH_MAX_PERIOD+i]);
  }
  /* Cross-correlation on half-frames. */
  for (sub=0;sub<2;sub++) {
    int off = sub*FRAME_SIZE/2;
    celt_pitch_xcorr(&st->exc_buf[PITCH_MAX_PERIOD+off], st->exc_buf+off, xcorr, FRAME_SIZE/2, PITCH_MAX_PERIOD);
    ener0 = celt_inner_prod(&st->exc_buf[PITCH_MAX_PERIOD+off], &st->exc_buf[PITCH_MAX_PERIOD+off], FRAME_SIZE/2);
    st->frame_weight[2+2*st->pcount+sub] = ener0;
    //printf("%f\n", st->frame_weight[2+2*st->pcount+sub]);
    for (i=0;i<PITCH_MAX_PERIOD;i++) {
      ener = (1 + ener0 + celt_inner_prod(&st->exc_buf[i+off], &st->exc_buf[i+off], FRAME_SIZE/2));
      st->xc[2+2*st->pcount+sub][i] = 2*xcorr[i] / ener;
    }

I can't understand the the st->pitch_filt =sum, why it can get pitch ???