dec_amr.cpp

实现3GPP的GSM中AMR语音的CODECS。

 HEAP MEMORY USED: x bytes

 STACK MEMORY USED: x bytes

 CLOCK CYCLES: (cycle count equation for this function) + (variable
                used to represent cycle count for each subroutine
                called)
     where: (cycle count variable) = cycle count for [subroutine
                                     name]

------------------------------------------------------------------------------
 CAUTION [optional]
 [State any special notes, constraints or cautions for users of this function]

------------------------------------------------------------------------------
*/

Word16 Decoder_amr_reset(Decoder_amrState *state, enum Mode mode)
{
    Word16 i;

    if (state == (Decoder_amrState *) NULL)
    {
        /* fprint(stderr, "Decoder_amr_reset: invalid parameter\n");  */
        return(-1);
    }

    /* Initialize static pointer */
    state->exc = state->old_exc + PIT_MAX + L_INTERPOL;

    /* Static vectors to zero */
    oscl_memset(state->old_exc, 0, sizeof(Word16)*(PIT_MAX + L_INTERPOL));

    if (mode != MRDTX)
    {
        oscl_memset(state->mem_syn, 0, sizeof(Word16)*M);
    }
    /* initialize pitch sharpening */
    state->sharp = SHARPMIN;
    state->old_T0 = 40;

    /* Initialize overflow Flag */

    state->overflow = 0;

    /* Initialize state->lsp_old [] */

    if (mode != MRDTX)
    {
        state->lsp_old[0] = 30000;
        state->lsp_old[1] = 26000;
        state->lsp_old[2] = 21000;
        state->lsp_old[3] = 15000;
        state->lsp_old[4] = 8000;
        state->lsp_old[5] = 0;
        state->lsp_old[6] = -8000;
        state->lsp_old[7] = -15000;
        state->lsp_old[8] = -21000;
        state->lsp_old[9] = -26000;
    }

    /* Initialize memories of bad frame handling */
    state->prev_bf = 0;
    state->prev_pdf = 0;
    state->state = 0;

    state->T0_lagBuff = 40;
    state->inBackgroundNoise = 0;
    state->voicedHangover = 0;
    if (mode != MRDTX)
    {
        for (i = 0;i < EXC_ENERGY_HIST_LEN;i++)
        {
            state->excEnergyHist[i] = 0;
        }
    }

    for (i = 0; i < LTP_GAIN_HISTORY_LEN; i++)
    {
        state->ltpGainHistory[i] = 0;
    }

    Cb_gain_average_reset(&(state->Cb_gain_averState));
    if (mode != MRDTX)
    {
        lsp_avg_reset(&(state->lsp_avg_st));
    }
    D_plsf_reset(&(state->lsfState));
    ec_gain_pitch_reset(&(state->ec_gain_p_st));
    ec_gain_code_reset(&(state->ec_gain_c_st));

    if (mode != MRDTX)
    {
        gc_pred_reset(&(state->pred_state));
    }

    Bgn_scd_reset(&(state->background_state));
    state->nodataSeed = 21845;
    ph_disp_reset(&(state->ph_disp_st));
    if (mode != MRDTX)
    {
        dtx_dec_reset(&(state->dtxDecoderState));
    }

    return(0);
}

/****************************************************************************/

/*
------------------------------------------------------------------------------
 FUNCTION NAME: Decoder_amr
------------------------------------------------------------------------------
 INPUT AND OUTPUT DEFINITIONS

 Inputs:
    st = pointer to a structure of type Decoder_amrState
    mode = codec mode (enum Mode)
    parm = buffer of synthesis parameters (Word16)
    frame_type = received frame type (enum RXFrameType)
    synth = buffer containing synthetic speech (Word16)
    A_t = buffer containing decoded LP filter in 4 subframes (Word16)

 Outputs:
    structure pointed to by st contains the newly calculated decoder
      parameters
    synth buffer contains the decoded speech samples
    A_t buffer contains the decoded LP filter parameters

 Returns:
    return_value = 0 (int)

 Global Variables Used:
    None

 Local Variables Needed:
    None

------------------------------------------------------------------------------
 FUNCTION DESCRIPTION

 This function performs the decoding of one speech frame for a given codec
 mode.

------------------------------------------------------------------------------
 REQUIREMENTS

 None

------------------------------------------------------------------------------
 REFERENCES

 dec_amr.c, UMTS GSM AMR speech codec, R99 - Version 3.2.0, March 2, 2001

------------------------------------------------------------------------------
 PSEUDO-CODE

int Decoder_amr (
    Decoder_amrState *st,      // i/o : State variables
    enum Mode mode,            // i   : AMR mode
    Word16 parm[],             // i   : vector of synthesis parameters
                                        (PRM_SIZE)
    enum RXFrameType frame_type, // i   : received frame type
    Word16 synth[],            // o   : synthesis speech (L_FRAME)
    Word16 A_t[]               // o   : decoded LP filter in 4 subframes
                                        (AZ_SIZE)
)
{
    // LPC coefficients

    Word16 *Az;                // Pointer on A_t

    // LSPs

    Word16 lsp_new[M];
    Word16 lsp_mid[M];

    // LSFs

    Word16 prev_lsf[M];
    Word16 lsf_i[M];

    // Algebraic codevector

    Word16 code[L_SUBFR];

    // excitation

    Word16 excp[L_SUBFR];
    Word16 exc_enhanced[L_SUBFR];

    // Scalars

    Word16 i, i_subfr;
    Word16 T0, T0_frac, index, index_mr475 = 0;
    Word16 gain_pit, gain_code, gain_code_mix, pit_sharp, pit_flag, pitch_fac;
    Word16 t0_min, t0_max;
    Word16 delta_frc_low, delta_frc_range;
    Word16 tmp_shift;
    Word16 temp;
    Word32 L_temp;
    Word16 flag4;
    Word16 carefulFlag;
    Word16 excEnergy;
    Word16 subfrNr;
    Word16 evenSubfr = 0;

    Word16 bfi = 0;   // bad frame indication flag
    Word16 pdfi = 0;  // potential degraded bad frame flag

    enum DTXStateType newDTXState;  // SPEECH , DTX, DTX_MUTE

    // find the new  DTX state  SPEECH OR DTX
    newDTXState = rx_dtx_handler(st->dtxDecoderState, frame_type);

    // DTX actions
    if (sub(newDTXState, SPEECH) != 0 )
    {
       Decoder_amr_reset (st, MRDTX);

       dtx_dec(st->dtxDecoderState,
               st->mem_syn,
               st->lsfState,
               st->pred_state,
               st->Cb_gain_averState,
               newDTXState,
               mode,
               parm, synth, A_t);
       // update average lsp

       Lsf_lsp(st->lsfState->past_lsf_q, st->lsp_old, M);
       lsp_avg(st->lsp_avg_st, st->lsfState->past_lsf_q);
       goto the_end;
    }

    // SPEECH action state machine
    if ((sub(frame_type, RX_SPEECH_BAD) == 0) ||
        (sub(frame_type, RX_NO_DATA) == 0) ||
        (sub(frame_type, RX_ONSET) == 0))
    {
       bfi = 1;
       if ((sub(frame_type, RX_NO_DATA) == 0) ||
           (sub(frame_type, RX_ONSET) == 0))
       {
	  build_CN_param(&st->nodataSeed,
			 prmno[mode],
			 bitno[mode],
			 parm);
       }
    }
    else if (sub(frame_type, RX_SPEECH_DEGRADED) == 0)
    {
       pdfi = 1;
    }

    if (bfi != 0)
    {
        st->state = add (st->state, 1);
    }
    else if (sub (st->state, 6) == 0)

    {
        st->state = 5;
    }
    else
    {
        st->state = 0;
    }

    if (sub (st->state, 6) > 0)
    {
        st->state = 6;
    }

    // If this frame is the first speech frame after CNI period,
    // set the BFH state machine to an appropriate state depending
    // on whether there was DTX muting before start of speech or not
    // If there was DTX muting, the first speech frame is muted.
    // If there was no DTX muting, the first speech frame is not
    // muted. The BFH state machine starts from state 5, however, to
    // keep the audible noise resulting from a SID frame which is
    // erroneously interpreted as a good speech frame as small as
    // possible (the decoder output in this case is quickly muted)

    if (sub(st->dtxDecoderState->dtxGlobalState, DTX) == 0)
    {
       st->state = 5;
       st->prev_bf = 0;
    }
    else if (sub(st->dtxDecoderState->dtxGlobalState, DTX_MUTE) == 0)
    {
       st->state = 5;
       st->prev_bf = 1;
    }

    // save old LSFs for CB gain smoothing
    Copy (st->lsfState->past_lsf_q, prev_lsf, M);

    // decode LSF parameters and generate interpolated lpc coefficients
       for the 4 subframes
    if (sub (mode, MR122) != 0)
    {
       D_plsf_3(st->lsfState, mode, bfi, parm, lsp_new);

       // Advance synthesis parameters pointer
       parm += 3;

       Int_lpc_1to3(st->lsp_old, lsp_new, A_t);
    }
    else
    {
       D_plsf_5 (st->lsfState, bfi, parm, lsp_mid, lsp_new);

       // Advance synthesis parameters pointer
       parm += 5;

       Int_lpc_1and3 (st->lsp_old, lsp_mid, lsp_new, A_t);
    }

    // update the LSPs for the next frame
    for (i = 0; i < M; i++)
    {
       st->lsp_old[i] = lsp_new[i];
    }

    *------------------------------------------------------------------------*
    *          Loop for every subframe in the analysis frame                 *
    *------------------------------------------------------------------------*
    * The subframe size is L_SUBFR and the loop is repeated L_FRAME/L_SUBFR  *
    *  times                                                                 *
    *     - decode the pitch delay                                           *
    *     - decode algebraic code                                            *
    *     - decode pitch and codebook gains                                  *
    *     - find the excitation and compute synthesis speech                 *
    *------------------------------------------------------------------------*

    // pointer to interpolated LPC parameters
    Az = A_t;

    evenSubfr = 0;
    subfrNr = -1;
    for (i_subfr = 0; i_subfr < L_FRAME; i_subfr += L_SUBFR)
    {
       subfrNr = add(subfrNr, 1);
       evenSubfr = sub(1, evenSubfr);

       // flag for first and 3th subframe
       pit_flag = i_subfr;

       if (sub (i_subfr, L_FRAME_BY2) == 0)
       {
          if (sub(mode, MR475) != 0 && sub(mode, MR515) != 0)
          {
             pit_flag = 0;
          }
       }

       // pitch index
       index = *parm++;

        *-------------------------------------------------------*
        * - decode pitch lag and find adaptive codebook vector. *
        *-------------------------------------------------------*

       if (sub(mode, MR122) != 0)
       {
          // flag4 indicates encoding with 4 bit resolution;
          // this is needed for mode MR475, MR515, MR59 and MR67

          flag4 = 0;
          if ((sub (mode, MR475) == 0) ||
              (sub (mode, MR515) == 0) ||
              (sub (mode, MR59) == 0) ||
              (sub (mode, MR67) == 0) ) {
             flag4 = 1;
          }

           *-------------------------------------------------------*
           * - get ranges for the t0_min and t0_max                *
           * - only needed in delta decoding                       *
           *-------------------------------------------------------*

          delta_frc_low = 5;
          delta_frc_range = 9;

          if ( sub(mode, MR795) == 0 )
          {
             delta_frc_low = 10;
             delta_frc_range = 19;
          }

          t0_min = sub(st->old_T0, delta_frc_low);
          if (sub(t0_min, PIT_MIN) < 0)
          {
             t0_min = PIT_MIN;
          }
          t0_max = add(t0_min, delta_frc_range);
          if (sub(t0_max, PIT_MAX) > 0)
          {
             t0_max = PIT_MAX;
             t0_min = sub(t0_max, delta_frc_range);
          }

          Dec_lag3 (index, t0_min, t0_max, pit_flag, st->old_T0,
                    &T0, &T0_frac, flag4);

          st->T0_lagBuff = T0;

          if (bfi != 0)
          {
             if (sub (st->old_T0, PIT_MAX) < 0)
             {                                      // Graceful pitch
                st->old_T0 = add(st->old_T0, 1);    // degradation
             }
             T0 = st->old_T0;
             T0_frac = 0;

             if ( st->inBackgroundNoise != 0 &&
                  sub(st->voicedHangover, 4) > 0 &&
                  ((sub(mode, MR475) == 0 ) ||
                   (sub(mode, MR515) == 0 ) ||
                   (sub(mode, MR59) == 0) )
                  )
             {
                T0 = st->T0_lagBuff;
             }
          }

          Pred_lt_3or6 (st->exc, T0, T0_frac, L_SUBFR, 1);
       }
       else