sp_enc.c

ffmpeg源码分析

}


/*
 * lsp
 *
 *
 * Parameters:
 *    req_mode          I: requested mode
 *    used_mode         I: used mode
 *    lsp_old           B: old LSP vector
 *    lsp_old_q         B: old quantized LSP vector
 *    past_rq           B: past quantized residual
 *    az                B: interpolated LP parameters
 *    azQ               O: quantization interpol. LP parameters
 *    lsp_new           O: new lsp vector
 *    anap              O: analysis parameters
 *
 * Function:
 *    From A(z) to lsp. LSP quantization and interpolation
 *
 * Returns:
 *    void
 */
static void lsp( enum Mode req_mode, enum Mode used_mode, Float32 *lsp_old,
      Float32 *lsp_old_q, Float32 *past_rq, Float32 az[], Float32 azQ[], Float32
      lsp_new[], Word16 **anap )
{
   Float32 lsp_new_q[M];   /* LSPs at 4th subframe */
   Float32 lsp_mid[M], lsp_mid_q[M];   /* LSPs at 2nd subframe */
   Word32 pred_init_i;   /* init index for MA prediction in DTX mode */


   if ( req_mode == MR122 ) {
      Az_lsp( &az[MP1], lsp_mid, lsp_old );
      Az_lsp( &az[MP1 * 3], lsp_new, lsp_mid );

      /*
       * Find interpolated LPC parameters in all subframes
       * (both quantized and unquantized).
       * The interpolated parameters are in array A_t[] of size (M+1)*4
       * and the quantized interpolated parameters are in array Aq_t[]
       */
      Int_lpc_1and3_2( lsp_old, lsp_mid, lsp_new, az );

      if ( used_mode != MRDTX ) {
         /* LSP quantization (lsp_mid[] and lsp_new[] jointly quantized) */
         Q_plsf_5( past_rq, lsp_mid, lsp_new, lsp_mid_q, lsp_new_q, *anap );
         Int_lpc_1and3( lsp_old_q, lsp_mid_q, lsp_new_q, azQ );

         /* Advance analysis parameters pointer */
         ( *anap ) += 5;
      }
   }
   else {
      /* From A(z) to lsp */
      Az_lsp( &az[MP1 * 3], lsp_new, lsp_old );

      /*
       * Find interpolated LPC parameters in all subframes
       * (both quantized and unquantized).
       * The interpolated parameters are in array A_t[] of size (M+1)*4
       * and the quantized interpolated parameters are in array Aq_t[]
       */
      Int_lpc_1to3_2( lsp_old, lsp_new, az );

      /* LSP quantization */
      if ( used_mode != MRDTX ) {
         Q_plsf_3( req_mode, past_rq, lsp_new, lsp_new_q, *anap, &pred_init_i );
         Int_lpc_1to3( lsp_old_q, lsp_new_q, azQ );

         /* Advance analysis parameters pointer */
         ( *anap ) += 3;
      }
   }

   /* update the LSPs for the next frame */
   memcpy( lsp_old, lsp_new, M <<2 );
   memcpy( lsp_old_q, lsp_new_q, M <<2 );
}


/*
 * check_lsp
 *
 *
 * Parameters:
 *    count          B: counter for resonance
 *    lsp            B: LSP vector
 *
 * Function:
 *    Check the LSP's to detect resonances
 *
 *    Resonances in the LPC filter are monitored to detect possible problem
 *    areas where divergence between the adaptive codebook memories in
 *    the encoder and the decoder could cause unstable filters in areas
 *    with highly correlated continuos signals. Typically, this divergence
 *    is due to channel errors.
 *    The monitoring of resonance signals is performed using unquantized LSPs
 *    q(i), i = 1,...,10. The algorithm utilises the fact that LSPs are
 *    closely located at a peak in the spectrum. First, two distances,
 *    dist 1 and dist 2 ,are calculated in two different regions,
 *    defined as
 *
 *    dist1 = min[q(i) - q(i + 1)],  i = 4,...,8
 *    dist2 = min[q(i) - q(i + 1)],  i = 2,3
 *
 *    Either of these two minimum distance conditions must be fulfilled
 *    to classify the frame as a resonance frame and increase the resonance
 *    counter.
 *
 *    if(dist1 < TH1) || if (dist2 < TH2)
 *       counter++
 *    else
 *       counter = 0
 *
 *    TH1 = 0.046
 *    TH2 = 0.018, q(2) > 0.98
 *    TH2 = 0.024, 0.93 < q(2) <= 0.98
 *    TH2 = 0.018, otherwise
 *
 *    12 consecutive resonance frames are needed to indicate possible
 *    problem conditions, otherwise the LSP_flag is cleared.
 *
 * Returns:
 *    resonance flag
 */
static Word16 check_lsp( Word16 *count, Float32 *lsp )
{
   Float32 dist, dist_min1, dist_min2, dist_th;
   Word32 i;


   /*
    * Check for a resonance:
    * Find minimum distance between lsp[i] and lsp[i+1]
    */
   dist_min1 = FLT_MAX;

   for ( i = 3; i < 8; i++ ) {
      dist = lsp[i] - lsp[i + 1];

      if ( dist < dist_min1 ) {
         dist_min1 = dist;
      }
   }
   dist_min2 = FLT_MAX;

   for ( i = 1; i < 3; i++ ) {
      dist = lsp[i] - lsp[i + 1];

      if ( dist < dist_min2 ) {
         dist_min2 = dist;
      }
   }

   if ( lsp[1] > 0.98F ) {
      dist_th = 0.018F;
   }
   else if ( lsp[1] > 0.93F ) {
      dist_th = 0.024F;
   }
   else {
      dist_th = 0.034F;
   }

   if ( ( dist_min1 < 0.046F ) || ( dist_min2 < dist_th ) ) {
      *count += 1;
   }
   else {
      *count = 0;
   }

   /* Need 12 consecutive frames to set the flag */
   if ( *count >= 12 ) {
      *count = 12;
      return 1;
   }
   else {
      return 0;
   }
}


/*
 * Weight_Ai
 *
 *
 * Parameters:
 *    a                 I: LPC coefficients                    [M+1]
 *    fac               I: Spectral expansion factors.         [M+1]
 *    a_exp             O: Spectral expanded LPC coefficients  [M+1]
 *
 * Function:
 *    Spectral expansion of LP coefficients
 *
 * Returns:
 *    void
 */
static void Weight_Ai( Float32 a[], const Float32 fac[], Float32 a_exp[] )
{
   Word32 i;


   a_exp[0] = a[0];

   for ( i = 1; i <= M; i++ ) {
      a_exp[i] = a[i] * fac[i - 1];
   }
   return;
}


/*
 * Residu
 *
 *
 * Parameters:
 *    a                 I: prediction coefficients
 *    x                 I: speech signal
 *    y                 O: residual signal
 *
 * Function:
 *    Computes the LTP residual signal.
 *
 * Returns:
 *    void
 */
static void Residu( Float32 a[], Float32 x[], Float32 y[] )
{
   Float32 s;
   Word32 i;


   for ( i = 0; i < L_SUBFR; i += 4 ) {
      s = x[i] * a[0];
      s += x[i - 1] *a[1];
      s += x[i - 2] * a[2];
      s += x[i - 3] * a[3];
      s += x[i - 4] * a[4];
      s += x[i - 5] * a[5];
      s += x[i - 6] * a[6];
      s += x[i - 7] * a[7];
      s += x[i - 8] * a[8];
      s += x[i - 9] * a[9];
      s += x[i - 10] * a[10];
      y[i] = s;
      s = x[i + 1] *a[0];
      s += x[i] * a[1];
      s += x[i - 1] *a[2];
      s += x[i - 2] * a[3];
      s += x[i - 3] * a[4];
      s += x[i - 4] * a[5];
      s += x[i - 5] * a[6];
      s += x[i - 6] * a[7];
      s += x[i - 7] * a[8];
      s += x[i - 8] * a[9];
      s += x[i - 9] * a[10];
      y[i + 1] = s;
      s = x[i + 2] * a[0];
      s += x[i + 1] *a[1];
      s += x[i] * a[2];
      s += x[i - 1] *a[3];
      s += x[i - 2] * a[4];
      s += x[i - 3] * a[5];
      s += x[i - 4] * a[6];
      s += x[i - 5] * a[7];
      s += x[i - 6] * a[8];
      s += x[i - 7] * a[9];
      s += x[i - 8] * a[10];
      y[i + 2] = s;
      s = x[i + 3] * a[0];
      s += x[i + 2] * a[1];
      s += x[i + 1] *a[2];
      s += x[i] * a[3];
      s += x[i - 1] *a[4];
      s += x[i - 2] * a[5];
      s += x[i - 3] * a[6];
      s += x[i - 4] * a[7];
      s += x[i - 5] * a[8];
      s += x[i - 6] * a[9];
      s += x[i - 7] * a[10];
      y[i + 3] = s;
   }
   return;
}


/*
 * Syn_filt
 *
 *
 * Parameters:
 *    a                 I: prediction coefficients [M+1]
 *    x                 I: input signal
 *    y                 O: output signal
 *    mem               B: memory associated with this filtering
 *    update            I: 0=no update, 1=update of memory.
 *
 * Function:
 *    Perform synthesis filtering through 1/A(z).
 *
 * Returns:
 *    void
 */
static void Syn_filt( Float32 a[], Float32 x[], Float32 y[], Float32 mem[],
      Word16 update )
{
   Float64 tmp[50];
   Float64 sum;
   Float64 *yy;
   Word32 i;


   /* Copy mem[] to yy[] */
   yy = tmp;

   for ( i = 0; i < M; i++ ) {
      *yy++ = mem[i];
   }

   /* Do the filtering. */
   for ( i = 0; i < L_SUBFR; i = i + 4 ) {
      sum = x[i] * a[0];
      sum -= a[1] * yy[ - 1];
      sum -= a[2] * yy[ - 2];
      sum -= a[3] * yy[ - 3];
      sum -= a[4] * yy[ - 4];
      sum -= a[5] * yy[ - 5];
      sum -= a[6] * yy[ - 6];
      sum -= a[7] * yy[ - 7];
      sum -= a[8] * yy[ - 8];
      sum -= a[9] * yy[ - 9];
      sum -= a[10] * yy[ - 10];
      *yy++ = sum;
      y[i] = ( Float32 )yy[ - 1];
      sum = x[i + 1] *a[0];
      sum -= a[1] * yy[ - 1];
      sum -= a[2] * yy[ - 2];
      sum -= a[3] * yy[ - 3];
      sum -= a[4] * yy[ - 4];
      sum -= a[5] * yy[ - 5];
      sum -= a[6] * yy[ - 6];
      sum -= a[7] * yy[ - 7];
      sum -= a[8] * yy[ - 8];
      sum -= a[9] * yy[ - 9];
      sum -= a[10] * yy[ - 10];
      *yy++ = sum;
      y[i + 1] = ( Float32 )yy[ - 1];
      sum = x[i + 2] * a[0];
      sum -= a[1] * yy[ - 1];
      sum -= a[2] * yy[ - 2];
      sum -= a[3] * yy[ - 3];
      sum -= a[4] * yy[ - 4];
      sum -= a[5] * yy[ - 5];
      sum -= a[6] * yy[ - 6];
      sum -= a[7] * yy[ - 7];
      sum -= a[8] * yy[ - 8];
      sum -= a[9] * yy[ - 9];
      sum -= a[10] * yy[ - 10];
      *yy++ = sum;
      y[i + 2] = ( Float32 )yy[ - 1];
      sum = x[i + 3] * a[0];
      sum -= a[1] * yy[ - 1];
      sum -= a[2] * yy[ - 2];
      sum -= a[3] * yy[ - 3];
      sum -= a[4] * yy[ - 4];
      sum -= a[5] * yy[ - 5];
      sum -= a[6] * yy[ - 6];
      sum -= a[7] * yy[ - 7];
      sum -= a[8] * yy[ - 8];
      sum -= a[9] * yy[ - 9];
      sum -= a[10] * yy[ - 10];
      *yy++ = sum;
      y[i + 3] = ( Float32 )yy[ - 1];
   }

   /* Update of memory if update==1 */
   if ( update != 0 ) {
      for ( i = 0; i < M; i++ ) {
         mem[i] = y[30 + i];
      }
   }
   return;
}


/*
 * pre_big
 *
 *
 * Parameters:
 *    mode              I: AMR mode
 *    gamma1            I: spectral exp. factor 1
 *    gamma1_12k2       I: spectral exp. factor 1 for modes above MR795
 *    gamma2            I: spectral exp. factor 2
 *    A_t               I: A(z) unquantized, for 4 subframes
 *    frame_offset      I: frameoffset, 1st or second big_sbf
 *    speech            I: speech
 *    mem_w             B: synthesis filter memory state
 *    wsp               O: weighted speech
 *
 * Function:
 *    Big subframe (2 subframes) preprocessing
 *
 *    Open-loop pitch analysis is performed in order to simplify the pitch
 *    analysis and confine the closed-loop pitch search to a small number of
 *    lags around the open-loop estimated lags.
 *    Open-loop pitch estimation is based on the weighted speech signal Sw(n)
 *    which is obtained by filtering the input speech signal through
 *    the weighting filter
 *
 *    W(z) = A(z/g1) / A(z/g2)
 *
 *    That is, in a subframe of size L, the weighted speech is given by:
 *
 *                    10                           10
 *    Sw(n) = S(n) + SUM[a(i) * g1(i) * S(n-i)] - SUM[a(i) * g2(i) * Sw(n-i)],
 *                   i=1                          i=1
 *    n = 0, ..., L-1
 *
 * Returns:
 *    void
 */
static Word32 pre_big( enum Mode mode, const Float32 gamma1[], const Float32
      gamma1_12k2[], const Float32 gamma2[], Float32 A_t[], Word16 frame_offset,
      Float32 speech[], Float32 mem_w[], Float32 wsp[] )
{
   Float32 Ap1[MP1], Ap2[MP1];
   Word32 offset, i;


   /* A(z) with spectral expansion */
   const Float32 *g1;


   g1 = gamma1_12k2;

   if ( mode <= MR795 ) {
      g1 = gamma1;
   }
   offset = 0;

   if ( frame_offset > 0 ) {
      offset = MP1 << 1;
   }

   /* process two subframes (which form the "big" subframe) */
   for ( i = 0; i < 2; i++ ) {
      /* a(i) * g1(i) */
      Weight_Ai( &A_t[offset], g1, Ap1 );

      /* a(i) * g2(i) */
      Weight_Ai( &A_t[offset], gamma2, Ap2 );

      /*
       *       10
       *  S(n) + SUM[a(i) * g1(i) * S(n-i)]
       *       i=1
       */
      Residu( Ap1, &speech[frame_offset], &wsp[frame_offset] );

      /*
       *          10                            10
       *  S(n) + SUM[a(i) * g1(i) * S(n-i)]    SUM[a(i) * g2(i) * Sn(n-i)]
       *         i=1                           i=1
       */
      Syn_filt( Ap2, &wsp[frame_offset], &wsp[frame_offset], mem_w, 1 );
      offset += MP1;
      frame_offset += L_SUBFR;
   }
   return 0;
}


/*
 * comp_corr
 *
 *
 * Parameters:
 *    sig               I: signal
 *    L_frame           I: length of frame to compute pitch
 *    lag_max           I: maximum lag
 *    lag_min           I: minimum lag
 *    corr              O: correlation of selected lag
 *
 * Function:
 *    Calculate all correlations in a given delay range.
 *
 * Returns:
 *    void
 */
static void comp_corr( Float32 sig[], Word32 L_frame, Word32 lag_max, Word32
      lag_min, Float32 corr[] )
{
   Word32 i, j;
   Float32 *p, *p1;
   Float32 T0;


   for ( i = lag_max; i >= lag_min; i-- ) {
      p = sig;
      p1 = &sig[ - i];
      T0 = 0.0F;

      for ( j = 0; j < L_frame; j = j + 40, p += 40, p1 += 40 ) {
         T0 += p[0] * p1[0] + p[1] * p1[1] + p[2] * p1[2] + p[3] * p1[3];
         T0 += p[4] * p1[4] + p[5] * p1[5] + p[6] * p1[6] + p[7] * p1[7];
         T0 += p[8] * p1[8] + p[9] * p1[9] + p[10] * p1[10] + p[11] * p1[11];
         T0 += p[12] * p1[12] + p[13] * p1[13] + p[14] * p1[14] + p[15] * p1[15]
         ;
         T0 += p[16] * p1[16] + p[17] * p1[17] + p[18] * p1[18] + p[19] * p1[19]
         ;
         T0 += p[20] * p1[20] + p[21] * p1[21] + p[22] * p1[22] + p[23] * p1[23]
         ;
         T0 += p[24] * p1[24] + p[25] * p1[25] + p[26] * p1[26] + p[27] * p1[27]
         ;
         T0 += p[28] * p1[28] + p[29] * p1[29] + p[30] * p1[30] + p[31] * p1[31]
         ;
         T0 += p[32] * p1[32] + p[33] * p1[33] + p[34] * p1[34] + p[35] * p1[35]
         ;
         T0 += p[36] * p1[36] + p[37] * p1[37] + p[38] * p1[38] + p[39] * p1[39]
         ;
      }
      corr[ - i] = T0;
   }
   return;
}