Subversion Repositories Kolibri OS

Rev

Go to most recent revision | Blame | Last modification | View Log | RSS feed

  1. /*
  2.  * AMR wideband decoder
  3.  * Copyright (c) 2010 Marcelo Galvao Povoa
  4.  *
  5.  * This file is part of FFmpeg.
  6.  *
  7.  * FFmpeg is free software; you can redistribute it and/or
  8.  * modify it under the terms of the GNU Lesser General Public
  9.  * License as published by the Free Software Foundation; either
  10.  * version 2.1 of the License, or (at your option) any later version.
  11.  *
  12.  * FFmpeg is distributed in the hope that it will be useful,
  13.  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14.  * MERCHANTABILITY or FITNESS FOR A particular PURPOSE.  See the GNU
  15.  * Lesser General Public License for more details.
  16.  *
  17.  * You should have received a copy of the GNU Lesser General Public
  18.  * License along with FFmpeg; if not, write to the Free Software
  19.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20.  */
  21.  
  22. /**
  23.  * @file
  24.  * AMR wideband decoder
  25.  */
  26.  
  27. #include "libavutil/channel_layout.h"
  28. #include "libavutil/common.h"
  29. #include "libavutil/float_dsp.h"
  30. #include "libavutil/lfg.h"
  31.  
  32. #include "avcodec.h"
  33. #include "lsp.h"
  34. #include "celp_filters.h"
  35. #include "celp_math.h"
  36. #include "acelp_filters.h"
  37. #include "acelp_vectors.h"
  38. #include "acelp_pitch_delay.h"
  39. #include "internal.h"
  40.  
  41. #define AMR_USE_16BIT_TABLES
  42. #include "amr.h"
  43.  
  44. #include "amrwbdata.h"
  45. #include "mips/amrwbdec_mips.h"
  46.  
  47. typedef struct {
  48.     AMRWBFrame                             frame; ///< AMRWB parameters decoded from bitstream
  49.     enum Mode                        fr_cur_mode; ///< mode index of current frame
  50.     uint8_t                           fr_quality; ///< frame quality index (FQI)
  51.     float                      isf_cur[LP_ORDER]; ///< working ISF vector from current frame
  52.     float                   isf_q_past[LP_ORDER]; ///< quantized ISF vector of the previous frame
  53.     float               isf_past_final[LP_ORDER]; ///< final processed ISF vector of the previous frame
  54.     double                      isp[4][LP_ORDER]; ///< ISP vectors from current frame
  55.     double               isp_sub4_past[LP_ORDER]; ///< ISP vector for the 4th subframe of the previous frame
  56.  
  57.     float                   lp_coef[4][LP_ORDER]; ///< Linear Prediction Coefficients from ISP vector
  58.  
  59.     uint8_t                       base_pitch_lag; ///< integer part of pitch lag for the next relative subframe
  60.     uint8_t                        pitch_lag_int; ///< integer part of pitch lag of the previous subframe
  61.  
  62.     float excitation_buf[AMRWB_P_DELAY_MAX + LP_ORDER + 2 + AMRWB_SFR_SIZE]; ///< current excitation and all necessary excitation history
  63.     float                            *excitation; ///< points to current excitation in excitation_buf[]
  64.  
  65.     float           pitch_vector[AMRWB_SFR_SIZE]; ///< adaptive codebook (pitch) vector for current subframe
  66.     float           fixed_vector[AMRWB_SFR_SIZE]; ///< algebraic codebook (fixed) vector for current subframe
  67.  
  68.     float                    prediction_error[4]; ///< quantified prediction errors {20log10(^gamma_gc)} for previous four subframes
  69.     float                          pitch_gain[6]; ///< quantified pitch gains for the current and previous five subframes
  70.     float                          fixed_gain[2]; ///< quantified fixed gains for the current and previous subframes
  71.  
  72.     float                              tilt_coef; ///< {beta_1} related to the voicing of the previous subframe
  73.  
  74.     float                 prev_sparse_fixed_gain; ///< previous fixed gain; used by anti-sparseness to determine "onset"
  75.     uint8_t                    prev_ir_filter_nr; ///< previous impulse response filter "impNr": 0 - strong, 1 - medium, 2 - none
  76.     float                           prev_tr_gain; ///< previous initial gain used by noise enhancer for threshold
  77.  
  78.     float samples_az[LP_ORDER + AMRWB_SFR_SIZE];         ///< low-band samples and memory from synthesis at 12.8kHz
  79.     float samples_up[UPS_MEM_SIZE + AMRWB_SFR_SIZE];     ///< low-band samples and memory processed for upsampling
  80.     float samples_hb[LP_ORDER_16k + AMRWB_SFR_SIZE_16k]; ///< high-band samples and memory from synthesis at 16kHz
  81.  
  82.     float          hpf_31_mem[2], hpf_400_mem[2]; ///< previous values in the high pass filters
  83.     float                           demph_mem[1]; ///< previous value in the de-emphasis filter
  84.     float               bpf_6_7_mem[HB_FIR_SIZE]; ///< previous values in the high-band band pass filter
  85.     float                 lpf_7_mem[HB_FIR_SIZE]; ///< previous values in the high-band low pass filter
  86.  
  87.     AVLFG                                   prng; ///< random number generator for white noise excitation
  88.     uint8_t                          first_frame; ///< flag active during decoding of the first frame
  89.     ACELPFContext                     acelpf_ctx; ///< context for filters for ACELP-based codecs
  90.     ACELPVContext                     acelpv_ctx; ///< context for vector operations for ACELP-based codecs
  91.     CELPFContext                       celpf_ctx; ///< context for filters for CELP-based codecs
  92.     CELPMContext                       celpm_ctx; ///< context for fixed point math operations
  93.  
  94. } AMRWBContext;
  95.  
  96. static av_cold int amrwb_decode_init(AVCodecContext *avctx)
  97. {
  98.     AMRWBContext *ctx = avctx->priv_data;
  99.     int i;
  100.  
  101.     if (avctx->channels > 1) {
  102.         avpriv_report_missing_feature(avctx, "multi-channel AMR");
  103.         return AVERROR_PATCHWELCOME;
  104.     }
  105.  
  106.     avctx->channels       = 1;
  107.     avctx->channel_layout = AV_CH_LAYOUT_MONO;
  108.     if (!avctx->sample_rate)
  109.         avctx->sample_rate = 16000;
  110.     avctx->sample_fmt     = AV_SAMPLE_FMT_FLT;
  111.  
  112.     av_lfg_init(&ctx->prng, 1);
  113.  
  114.     ctx->excitation  = &ctx->excitation_buf[AMRWB_P_DELAY_MAX + LP_ORDER + 1];
  115.     ctx->first_frame = 1;
  116.  
  117.     for (i = 0; i < LP_ORDER; i++)
  118.         ctx->isf_past_final[i] = isf_init[i] * (1.0f / (1 << 15));
  119.  
  120.     for (i = 0; i < 4; i++)
  121.         ctx->prediction_error[i] = MIN_ENERGY;
  122.  
  123.     ff_acelp_filter_init(&ctx->acelpf_ctx);
  124.     ff_acelp_vectors_init(&ctx->acelpv_ctx);
  125.     ff_celp_filter_init(&ctx->celpf_ctx);
  126.     ff_celp_math_init(&ctx->celpm_ctx);
  127.  
  128.     return 0;
  129. }
  130.  
  131. /**
  132.  * Decode the frame header in the "MIME/storage" format. This format
  133.  * is simpler and does not carry the auxiliary frame information.
  134.  *
  135.  * @param[in] ctx                  The Context
  136.  * @param[in] buf                  Pointer to the input buffer
  137.  *
  138.  * @return The decoded header length in bytes
  139.  */
  140. static int decode_mime_header(AMRWBContext *ctx, const uint8_t *buf)
  141. {
  142.     /* Decode frame header (1st octet) */
  143.     ctx->fr_cur_mode  = buf[0] >> 3 & 0x0F;
  144.     ctx->fr_quality   = (buf[0] & 0x4) == 0x4;
  145.  
  146.     return 1;
  147. }
  148.  
  149. /**
  150.  * Decode quantized ISF vectors using 36-bit indexes (6K60 mode only).
  151.  *
  152.  * @param[in]  ind                 Array of 5 indexes
  153.  * @param[out] isf_q               Buffer for isf_q[LP_ORDER]
  154.  *
  155.  */
  156. static void decode_isf_indices_36b(uint16_t *ind, float *isf_q)
  157. {
  158.     int i;
  159.  
  160.     for (i = 0; i < 9; i++)
  161.         isf_q[i]      = dico1_isf[ind[0]][i]      * (1.0f / (1 << 15));
  162.  
  163.     for (i = 0; i < 7; i++)
  164.         isf_q[i + 9]  = dico2_isf[ind[1]][i]      * (1.0f / (1 << 15));
  165.  
  166.     for (i = 0; i < 5; i++)
  167.         isf_q[i]     += dico21_isf_36b[ind[2]][i] * (1.0f / (1 << 15));
  168.  
  169.     for (i = 0; i < 4; i++)
  170.         isf_q[i + 5] += dico22_isf_36b[ind[3]][i] * (1.0f / (1 << 15));
  171.  
  172.     for (i = 0; i < 7; i++)
  173.         isf_q[i + 9] += dico23_isf_36b[ind[4]][i] * (1.0f / (1 << 15));
  174. }
  175.  
  176. /**
  177.  * Decode quantized ISF vectors using 46-bit indexes (except 6K60 mode).
  178.  *
  179.  * @param[in]  ind                 Array of 7 indexes
  180.  * @param[out] isf_q               Buffer for isf_q[LP_ORDER]
  181.  *
  182.  */
  183. static void decode_isf_indices_46b(uint16_t *ind, float *isf_q)
  184. {
  185.     int i;
  186.  
  187.     for (i = 0; i < 9; i++)
  188.         isf_q[i]       = dico1_isf[ind[0]][i]  * (1.0f / (1 << 15));
  189.  
  190.     for (i = 0; i < 7; i++)
  191.         isf_q[i + 9]   = dico2_isf[ind[1]][i]  * (1.0f / (1 << 15));
  192.  
  193.     for (i = 0; i < 3; i++)
  194.         isf_q[i]      += dico21_isf[ind[2]][i] * (1.0f / (1 << 15));
  195.  
  196.     for (i = 0; i < 3; i++)
  197.         isf_q[i + 3]  += dico22_isf[ind[3]][i] * (1.0f / (1 << 15));
  198.  
  199.     for (i = 0; i < 3; i++)
  200.         isf_q[i + 6]  += dico23_isf[ind[4]][i] * (1.0f / (1 << 15));
  201.  
  202.     for (i = 0; i < 3; i++)
  203.         isf_q[i + 9]  += dico24_isf[ind[5]][i] * (1.0f / (1 << 15));
  204.  
  205.     for (i = 0; i < 4; i++)
  206.         isf_q[i + 12] += dico25_isf[ind[6]][i] * (1.0f / (1 << 15));
  207. }
  208.  
  209. /**
  210.  * Apply mean and past ISF values using the prediction factor.
  211.  * Updates past ISF vector.
  212.  *
  213.  * @param[in,out] isf_q            Current quantized ISF
  214.  * @param[in,out] isf_past         Past quantized ISF
  215.  *
  216.  */
  217. static void isf_add_mean_and_past(float *isf_q, float *isf_past)
  218. {
  219.     int i;
  220.     float tmp;
  221.  
  222.     for (i = 0; i < LP_ORDER; i++) {
  223.         tmp = isf_q[i];
  224.         isf_q[i] += isf_mean[i] * (1.0f / (1 << 15));
  225.         isf_q[i] += PRED_FACTOR * isf_past[i];
  226.         isf_past[i] = tmp;
  227.     }
  228. }
  229.  
  230. /**
  231.  * Interpolate the fourth ISP vector from current and past frames
  232.  * to obtain an ISP vector for each subframe.
  233.  *
  234.  * @param[in,out] isp_q            ISPs for each subframe
  235.  * @param[in]     isp4_past        Past ISP for subframe 4
  236.  */
  237. static void interpolate_isp(double isp_q[4][LP_ORDER], const double *isp4_past)
  238. {
  239.     int i, k;
  240.  
  241.     for (k = 0; k < 3; k++) {
  242.         float c = isfp_inter[k];
  243.         for (i = 0; i < LP_ORDER; i++)
  244.             isp_q[k][i] = (1.0 - c) * isp4_past[i] + c * isp_q[3][i];
  245.     }
  246. }
  247.  
  248. /**
  249.  * Decode an adaptive codebook index into pitch lag (except 6k60, 8k85 modes).
  250.  * Calculate integer lag and fractional lag always using 1/4 resolution.
  251.  * In 1st and 3rd subframes the index is relative to last subframe integer lag.
  252.  *
  253.  * @param[out]    lag_int          Decoded integer pitch lag
  254.  * @param[out]    lag_frac         Decoded fractional pitch lag
  255.  * @param[in]     pitch_index      Adaptive codebook pitch index
  256.  * @param[in,out] base_lag_int     Base integer lag used in relative subframes
  257.  * @param[in]     subframe         Current subframe index (0 to 3)
  258.  */
  259. static void decode_pitch_lag_high(int *lag_int, int *lag_frac, int pitch_index,
  260.                                   uint8_t *base_lag_int, int subframe)
  261. {
  262.     if (subframe == 0 || subframe == 2) {
  263.         if (pitch_index < 376) {
  264.             *lag_int  = (pitch_index + 137) >> 2;
  265.             *lag_frac = pitch_index - (*lag_int << 2) + 136;
  266.         } else if (pitch_index < 440) {
  267.             *lag_int  = (pitch_index + 257 - 376) >> 1;
  268.             *lag_frac = (pitch_index - (*lag_int << 1) + 256 - 376) << 1;
  269.             /* the actual resolution is 1/2 but expressed as 1/4 */
  270.         } else {
  271.             *lag_int  = pitch_index - 280;
  272.             *lag_frac = 0;
  273.         }
  274.         /* minimum lag for next subframe */
  275.         *base_lag_int = av_clip(*lag_int - 8 - (*lag_frac < 0),
  276.                                 AMRWB_P_DELAY_MIN, AMRWB_P_DELAY_MAX - 15);
  277.         // XXX: the spec states clearly that *base_lag_int should be
  278.         // the nearest integer to *lag_int (minus 8), but the ref code
  279.         // actually always uses its floor, I'm following the latter
  280.     } else {
  281.         *lag_int  = (pitch_index + 1) >> 2;
  282.         *lag_frac = pitch_index - (*lag_int << 2);
  283.         *lag_int += *base_lag_int;
  284.     }
  285. }
  286.  
  287. /**
  288.  * Decode an adaptive codebook index into pitch lag for 8k85 and 6k60 modes.
  289.  * The description is analogous to decode_pitch_lag_high, but in 6k60 the
  290.  * relative index is used for all subframes except the first.
  291.  */
  292. static void decode_pitch_lag_low(int *lag_int, int *lag_frac, int pitch_index,
  293.                                  uint8_t *base_lag_int, int subframe, enum Mode mode)
  294. {
  295.     if (subframe == 0 || (subframe == 2 && mode != MODE_6k60)) {
  296.         if (pitch_index < 116) {
  297.             *lag_int  = (pitch_index + 69) >> 1;
  298.             *lag_frac = (pitch_index - (*lag_int << 1) + 68) << 1;
  299.         } else {
  300.             *lag_int  = pitch_index - 24;
  301.             *lag_frac = 0;
  302.         }
  303.         // XXX: same problem as before
  304.         *base_lag_int = av_clip(*lag_int - 8 - (*lag_frac < 0),
  305.                                 AMRWB_P_DELAY_MIN, AMRWB_P_DELAY_MAX - 15);
  306.     } else {
  307.         *lag_int  = (pitch_index + 1) >> 1;
  308.         *lag_frac = (pitch_index - (*lag_int << 1)) << 1;
  309.         *lag_int += *base_lag_int;
  310.     }
  311. }
  312.  
  313. /**
  314.  * Find the pitch vector by interpolating the past excitation at the
  315.  * pitch delay, which is obtained in this function.
  316.  *
  317.  * @param[in,out] ctx              The context
  318.  * @param[in]     amr_subframe     Current subframe data
  319.  * @param[in]     subframe         Current subframe index (0 to 3)
  320.  */
  321. static void decode_pitch_vector(AMRWBContext *ctx,
  322.                                 const AMRWBSubFrame *amr_subframe,
  323.                                 const int subframe)
  324. {
  325.     int pitch_lag_int, pitch_lag_frac;
  326.     int i;
  327.     float *exc     = ctx->excitation;
  328.     enum Mode mode = ctx->fr_cur_mode;
  329.  
  330.     if (mode <= MODE_8k85) {
  331.         decode_pitch_lag_low(&pitch_lag_int, &pitch_lag_frac, amr_subframe->adap,
  332.                               &ctx->base_pitch_lag, subframe, mode);
  333.     } else
  334.         decode_pitch_lag_high(&pitch_lag_int, &pitch_lag_frac, amr_subframe->adap,
  335.                               &ctx->base_pitch_lag, subframe);
  336.  
  337.     ctx->pitch_lag_int = pitch_lag_int;
  338.     pitch_lag_int += pitch_lag_frac > 0;
  339.  
  340.     /* Calculate the pitch vector by interpolating the past excitation at the
  341.        pitch lag using a hamming windowed sinc function */
  342.     ctx->acelpf_ctx.acelp_interpolatef(exc,
  343.                           exc + 1 - pitch_lag_int,
  344.                           ac_inter, 4,
  345.                           pitch_lag_frac + (pitch_lag_frac > 0 ? 0 : 4),
  346.                           LP_ORDER, AMRWB_SFR_SIZE + 1);
  347.  
  348.     /* Check which pitch signal path should be used
  349.      * 6k60 and 8k85 modes have the ltp flag set to 0 */
  350.     if (amr_subframe->ltp) {
  351.         memcpy(ctx->pitch_vector, exc, AMRWB_SFR_SIZE * sizeof(float));
  352.     } else {
  353.         for (i = 0; i < AMRWB_SFR_SIZE; i++)
  354.             ctx->pitch_vector[i] = 0.18 * exc[i - 1] + 0.64 * exc[i] +
  355.                                    0.18 * exc[i + 1];
  356.         memcpy(exc, ctx->pitch_vector, AMRWB_SFR_SIZE * sizeof(float));
  357.     }
  358. }
  359.  
  360. /** Get x bits in the index interval [lsb,lsb+len-1] inclusive */
  361. #define BIT_STR(x,lsb,len) (((x) >> (lsb)) & ((1 << (len)) - 1))
  362.  
  363. /** Get the bit at specified position */
  364. #define BIT_POS(x, p) (((x) >> (p)) & 1)
  365.  
  366. /**
  367.  * The next six functions decode_[i]p_track decode exactly i pulses
  368.  * positions and amplitudes (-1 or 1) in a subframe track using
  369.  * an encoded pulse indexing (TS 26.190 section 5.8.2).
  370.  *
  371.  * The results are given in out[], in which a negative number means
  372.  * amplitude -1 and vice versa (i.e., ampl(x) = x / abs(x) ).
  373.  *
  374.  * @param[out] out                 Output buffer (writes i elements)
  375.  * @param[in]  code                Pulse index (no. of bits varies, see below)
  376.  * @param[in]  m                   (log2) Number of potential positions
  377.  * @param[in]  off                 Offset for decoded positions
  378.  */
  379. static inline void decode_1p_track(int *out, int code, int m, int off)
  380. {
  381.     int pos = BIT_STR(code, 0, m) + off; ///code: m+1 bits
  382.  
  383.     out[0] = BIT_POS(code, m) ? -pos : pos;
  384. }
  385.  
  386. static inline void decode_2p_track(int *out, int code, int m, int off) ///code: 2m+1 bits
  387. {
  388.     int pos0 = BIT_STR(code, m, m) + off;
  389.     int pos1 = BIT_STR(code, 0, m) + off;
  390.  
  391.     out[0] = BIT_POS(code, 2*m) ? -pos0 : pos0;
  392.     out[1] = BIT_POS(code, 2*m) ? -pos1 : pos1;
  393.     out[1] = pos0 > pos1 ? -out[1] : out[1];
  394. }
  395.  
  396. static void decode_3p_track(int *out, int code, int m, int off) ///code: 3m+1 bits
  397. {
  398.     int half_2p = BIT_POS(code, 2*m - 1) << (m - 1);
  399.  
  400.     decode_2p_track(out, BIT_STR(code, 0, 2*m - 1),
  401.                     m - 1, off + half_2p);
  402.     decode_1p_track(out + 2, BIT_STR(code, 2*m, m + 1), m, off);
  403. }
  404.  
  405. static void decode_4p_track(int *out, int code, int m, int off) ///code: 4m bits
  406. {
  407.     int half_4p, subhalf_2p;
  408.     int b_offset = 1 << (m - 1);
  409.  
  410.     switch (BIT_STR(code, 4*m - 2, 2)) { /* case ID (2 bits) */
  411.     case 0: /* 0 pulses in A, 4 pulses in B or vice versa */
  412.         half_4p = BIT_POS(code, 4*m - 3) << (m - 1); // which has 4 pulses
  413.         subhalf_2p = BIT_POS(code, 2*m - 3) << (m - 2);
  414.  
  415.         decode_2p_track(out, BIT_STR(code, 0, 2*m - 3),
  416.                         m - 2, off + half_4p + subhalf_2p);
  417.         decode_2p_track(out + 2, BIT_STR(code, 2*m - 2, 2*m - 1),
  418.                         m - 1, off + half_4p);
  419.         break;
  420.     case 1: /* 1 pulse in A, 3 pulses in B */
  421.         decode_1p_track(out, BIT_STR(code,  3*m - 2, m),
  422.                         m - 1, off);
  423.         decode_3p_track(out + 1, BIT_STR(code, 0, 3*m - 2),
  424.                         m - 1, off + b_offset);
  425.         break;
  426.     case 2: /* 2 pulses in each half */
  427.         decode_2p_track(out, BIT_STR(code, 2*m - 1, 2*m - 1),
  428.                         m - 1, off);
  429.         decode_2p_track(out + 2, BIT_STR(code, 0, 2*m - 1),
  430.                         m - 1, off + b_offset);
  431.         break;
  432.     case 3: /* 3 pulses in A, 1 pulse in B */
  433.         decode_3p_track(out, BIT_STR(code, m, 3*m - 2),
  434.                         m - 1, off);
  435.         decode_1p_track(out + 3, BIT_STR(code, 0, m),
  436.                         m - 1, off + b_offset);
  437.         break;
  438.     }
  439. }
  440.  
  441. static void decode_5p_track(int *out, int code, int m, int off) ///code: 5m bits
  442. {
  443.     int half_3p = BIT_POS(code, 5*m - 1) << (m - 1);
  444.  
  445.     decode_3p_track(out, BIT_STR(code, 2*m + 1, 3*m - 2),
  446.                     m - 1, off + half_3p);
  447.  
  448.     decode_2p_track(out + 3, BIT_STR(code, 0, 2*m + 1), m, off);
  449. }
  450.  
  451. static void decode_6p_track(int *out, int code, int m, int off) ///code: 6m-2 bits
  452. {
  453.     int b_offset = 1 << (m - 1);
  454.     /* which half has more pulses in cases 0 to 2 */
  455.     int half_more  = BIT_POS(code, 6*m - 5) << (m - 1);
  456.     int half_other = b_offset - half_more;
  457.  
  458.     switch (BIT_STR(code, 6*m - 4, 2)) { /* case ID (2 bits) */
  459.     case 0: /* 0 pulses in A, 6 pulses in B or vice versa */
  460.         decode_1p_track(out, BIT_STR(code, 0, m),
  461.                         m - 1, off + half_more);
  462.         decode_5p_track(out + 1, BIT_STR(code, m, 5*m - 5),
  463.                         m - 1, off + half_more);
  464.         break;
  465.     case 1: /* 1 pulse in A, 5 pulses in B or vice versa */
  466.         decode_1p_track(out, BIT_STR(code, 0, m),
  467.                         m - 1, off + half_other);
  468.         decode_5p_track(out + 1, BIT_STR(code, m, 5*m - 5),
  469.                         m - 1, off + half_more);
  470.         break;
  471.     case 2: /* 2 pulses in A, 4 pulses in B or vice versa */
  472.         decode_2p_track(out, BIT_STR(code, 0, 2*m - 1),
  473.                         m - 1, off + half_other);
  474.         decode_4p_track(out + 2, BIT_STR(code, 2*m - 1, 4*m - 4),
  475.                         m - 1, off + half_more);
  476.         break;
  477.     case 3: /* 3 pulses in A, 3 pulses in B */
  478.         decode_3p_track(out, BIT_STR(code, 3*m - 2, 3*m - 2),
  479.                         m - 1, off);
  480.         decode_3p_track(out + 3, BIT_STR(code, 0, 3*m - 2),
  481.                         m - 1, off + b_offset);
  482.         break;
  483.     }
  484. }
  485.  
  486. /**
  487.  * Decode the algebraic codebook index to pulse positions and signs,
  488.  * then construct the algebraic codebook vector.
  489.  *
  490.  * @param[out] fixed_vector        Buffer for the fixed codebook excitation
  491.  * @param[in]  pulse_hi            MSBs part of the pulse index array (higher modes only)
  492.  * @param[in]  pulse_lo            LSBs part of the pulse index array
  493.  * @param[in]  mode                Mode of the current frame
  494.  */
  495. static void decode_fixed_vector(float *fixed_vector, const uint16_t *pulse_hi,
  496.                                 const uint16_t *pulse_lo, const enum Mode mode)
  497. {
  498.     /* sig_pos stores for each track the decoded pulse position indexes
  499.      * (1-based) multiplied by its corresponding amplitude (+1 or -1) */
  500.     int sig_pos[4][6];
  501.     int spacing = (mode == MODE_6k60) ? 2 : 4;
  502.     int i, j;
  503.  
  504.     switch (mode) {
  505.     case MODE_6k60:
  506.         for (i = 0; i < 2; i++)
  507.             decode_1p_track(sig_pos[i], pulse_lo[i], 5, 1);
  508.         break;
  509.     case MODE_8k85:
  510.         for (i = 0; i < 4; i++)
  511.             decode_1p_track(sig_pos[i], pulse_lo[i], 4, 1);
  512.         break;
  513.     case MODE_12k65:
  514.         for (i = 0; i < 4; i++)
  515.             decode_2p_track(sig_pos[i], pulse_lo[i], 4, 1);
  516.         break;
  517.     case MODE_14k25:
  518.         for (i = 0; i < 2; i++)
  519.             decode_3p_track(sig_pos[i], pulse_lo[i], 4, 1);
  520.         for (i = 2; i < 4; i++)
  521.             decode_2p_track(sig_pos[i], pulse_lo[i], 4, 1);
  522.         break;
  523.     case MODE_15k85:
  524.         for (i = 0; i < 4; i++)
  525.             decode_3p_track(sig_pos[i], pulse_lo[i], 4, 1);
  526.         break;
  527.     case MODE_18k25:
  528.         for (i = 0; i < 4; i++)
  529.             decode_4p_track(sig_pos[i], (int) pulse_lo[i] +
  530.                            ((int) pulse_hi[i] << 14), 4, 1);
  531.         break;
  532.     case MODE_19k85:
  533.         for (i = 0; i < 2; i++)
  534.             decode_5p_track(sig_pos[i], (int) pulse_lo[i] +
  535.                            ((int) pulse_hi[i] << 10), 4, 1);
  536.         for (i = 2; i < 4; i++)
  537.             decode_4p_track(sig_pos[i], (int) pulse_lo[i] +
  538.                            ((int) pulse_hi[i] << 14), 4, 1);
  539.         break;
  540.     case MODE_23k05:
  541.     case MODE_23k85:
  542.         for (i = 0; i < 4; i++)
  543.             decode_6p_track(sig_pos[i], (int) pulse_lo[i] +
  544.                            ((int) pulse_hi[i] << 11), 4, 1);
  545.         break;
  546.     }
  547.  
  548.     memset(fixed_vector, 0, sizeof(float) * AMRWB_SFR_SIZE);
  549.  
  550.     for (i = 0; i < 4; i++)
  551.         for (j = 0; j < pulses_nb_per_mode_tr[mode][i]; j++) {
  552.             int pos = (FFABS(sig_pos[i][j]) - 1) * spacing + i;
  553.  
  554.             fixed_vector[pos] += sig_pos[i][j] < 0 ? -1.0 : 1.0;
  555.         }
  556. }
  557.  
  558. /**
  559.  * Decode pitch gain and fixed gain correction factor.
  560.  *
  561.  * @param[in]  vq_gain             Vector-quantized index for gains
  562.  * @param[in]  mode                Mode of the current frame
  563.  * @param[out] fixed_gain_factor   Decoded fixed gain correction factor
  564.  * @param[out] pitch_gain          Decoded pitch gain
  565.  */
  566. static void decode_gains(const uint8_t vq_gain, const enum Mode mode,
  567.                          float *fixed_gain_factor, float *pitch_gain)
  568. {
  569.     const int16_t *gains = (mode <= MODE_8k85 ? qua_gain_6b[vq_gain] :
  570.                                                 qua_gain_7b[vq_gain]);
  571.  
  572.     *pitch_gain        = gains[0] * (1.0f / (1 << 14));
  573.     *fixed_gain_factor = gains[1] * (1.0f / (1 << 11));
  574. }
  575.  
  576. /**
  577.  * Apply pitch sharpening filters to the fixed codebook vector.
  578.  *
  579.  * @param[in]     ctx              The context
  580.  * @param[in,out] fixed_vector     Fixed codebook excitation
  581.  */
  582. // XXX: Spec states this procedure should be applied when the pitch
  583. // lag is less than 64, but this checking seems absent in reference and AMR-NB
  584. static void pitch_sharpening(AMRWBContext *ctx, float *fixed_vector)
  585. {
  586.     int i;
  587.  
  588.     /* Tilt part */
  589.     for (i = AMRWB_SFR_SIZE - 1; i != 0; i--)
  590.         fixed_vector[i] -= fixed_vector[i - 1] * ctx->tilt_coef;
  591.  
  592.     /* Periodicity enhancement part */
  593.     for (i = ctx->pitch_lag_int; i < AMRWB_SFR_SIZE; i++)
  594.         fixed_vector[i] += fixed_vector[i - ctx->pitch_lag_int] * 0.85;
  595. }
  596.  
  597. /**
  598.  * Calculate the voicing factor (-1.0 = unvoiced to 1.0 = voiced).
  599.  *
  600.  * @param[in] p_vector, f_vector   Pitch and fixed excitation vectors
  601.  * @param[in] p_gain, f_gain       Pitch and fixed gains
  602.  * @param[in] ctx                  The context
  603.  */
  604. // XXX: There is something wrong with the precision here! The magnitudes
  605. // of the energies are not correct. Please check the reference code carefully
  606. static float voice_factor(float *p_vector, float p_gain,
  607.                           float *f_vector, float f_gain,
  608.                           CELPMContext *ctx)
  609. {
  610.     double p_ener = (double) ctx->dot_productf(p_vector, p_vector,
  611.                                                           AMRWB_SFR_SIZE) *
  612.                     p_gain * p_gain;
  613.     double f_ener = (double) ctx->dot_productf(f_vector, f_vector,
  614.                                                           AMRWB_SFR_SIZE) *
  615.                     f_gain * f_gain;
  616.  
  617.     return (p_ener - f_ener) / (p_ener + f_ener);
  618. }
  619.  
  620. /**
  621.  * Reduce fixed vector sparseness by smoothing with one of three IR filters,
  622.  * also known as "adaptive phase dispersion".
  623.  *
  624.  * @param[in]     ctx              The context
  625.  * @param[in,out] fixed_vector     Unfiltered fixed vector
  626.  * @param[out]    buf              Space for modified vector if necessary
  627.  *
  628.  * @return The potentially overwritten filtered fixed vector address
  629.  */
  630. static float *anti_sparseness(AMRWBContext *ctx,
  631.                               float *fixed_vector, float *buf)
  632. {
  633.     int ir_filter_nr;
  634.  
  635.     if (ctx->fr_cur_mode > MODE_8k85) // no filtering in higher modes
  636.         return fixed_vector;
  637.  
  638.     if (ctx->pitch_gain[0] < 0.6) {
  639.         ir_filter_nr = 0;      // strong filtering
  640.     } else if (ctx->pitch_gain[0] < 0.9) {
  641.         ir_filter_nr = 1;      // medium filtering
  642.     } else
  643.         ir_filter_nr = 2;      // no filtering
  644.  
  645.     /* detect 'onset' */
  646.     if (ctx->fixed_gain[0] > 3.0 * ctx->fixed_gain[1]) {
  647.         if (ir_filter_nr < 2)
  648.             ir_filter_nr++;
  649.     } else {
  650.         int i, count = 0;
  651.  
  652.         for (i = 0; i < 6; i++)
  653.             if (ctx->pitch_gain[i] < 0.6)
  654.                 count++;
  655.  
  656.         if (count > 2)
  657.             ir_filter_nr = 0;
  658.  
  659.         if (ir_filter_nr > ctx->prev_ir_filter_nr + 1)
  660.             ir_filter_nr--;
  661.     }
  662.  
  663.     /* update ir filter strength history */
  664.     ctx->prev_ir_filter_nr = ir_filter_nr;
  665.  
  666.     ir_filter_nr += (ctx->fr_cur_mode == MODE_8k85);
  667.  
  668.     if (ir_filter_nr < 2) {
  669.         int i;
  670.         const float *coef = ir_filters_lookup[ir_filter_nr];
  671.  
  672.         /* Circular convolution code in the reference
  673.          * decoder was modified to avoid using one
  674.          * extra array. The filtered vector is given by:
  675.          *
  676.          * c2(n) = sum(i,0,len-1){ c(i) * coef( (n - i + len) % len ) }
  677.          */
  678.  
  679.         memset(buf, 0, sizeof(float) * AMRWB_SFR_SIZE);
  680.         for (i = 0; i < AMRWB_SFR_SIZE; i++)
  681.             if (fixed_vector[i])
  682.                 ff_celp_circ_addf(buf, buf, coef, i, fixed_vector[i],
  683.                                   AMRWB_SFR_SIZE);
  684.         fixed_vector = buf;
  685.     }
  686.  
  687.     return fixed_vector;
  688. }
  689.  
  690. /**
  691.  * Calculate a stability factor {teta} based on distance between
  692.  * current and past isf. A value of 1 shows maximum signal stability.
  693.  */
  694. static float stability_factor(const float *isf, const float *isf_past)
  695. {
  696.     int i;
  697.     float acc = 0.0;
  698.  
  699.     for (i = 0; i < LP_ORDER - 1; i++)
  700.         acc += (isf[i] - isf_past[i]) * (isf[i] - isf_past[i]);
  701.  
  702.     // XXX: This part is not so clear from the reference code
  703.     // the result is more accurate changing the "/ 256" to "* 512"
  704.     return FFMAX(0.0, 1.25 - acc * 0.8 * 512);
  705. }
  706.  
  707. /**
  708.  * Apply a non-linear fixed gain smoothing in order to reduce
  709.  * fluctuation in the energy of excitation.
  710.  *
  711.  * @param[in]     fixed_gain       Unsmoothed fixed gain
  712.  * @param[in,out] prev_tr_gain     Previous threshold gain (updated)
  713.  * @param[in]     voice_fac        Frame voicing factor
  714.  * @param[in]     stab_fac         Frame stability factor
  715.  *
  716.  * @return The smoothed gain
  717.  */
  718. static float noise_enhancer(float fixed_gain, float *prev_tr_gain,
  719.                             float voice_fac,  float stab_fac)
  720. {
  721.     float sm_fac = 0.5 * (1 - voice_fac) * stab_fac;
  722.     float g0;
  723.  
  724.     // XXX: the following fixed-point constants used to in(de)crement
  725.     // gain by 1.5dB were taken from the reference code, maybe it could
  726.     // be simpler
  727.     if (fixed_gain < *prev_tr_gain) {
  728.         g0 = FFMIN(*prev_tr_gain, fixed_gain + fixed_gain *
  729.                      (6226 * (1.0f / (1 << 15)))); // +1.5 dB
  730.     } else
  731.         g0 = FFMAX(*prev_tr_gain, fixed_gain *
  732.                     (27536 * (1.0f / (1 << 15)))); // -1.5 dB
  733.  
  734.     *prev_tr_gain = g0; // update next frame threshold
  735.  
  736.     return sm_fac * g0 + (1 - sm_fac) * fixed_gain;
  737. }
  738.  
  739. /**
  740.  * Filter the fixed_vector to emphasize the higher frequencies.
  741.  *
  742.  * @param[in,out] fixed_vector     Fixed codebook vector
  743.  * @param[in]     voice_fac        Frame voicing factor
  744.  */
  745. static void pitch_enhancer(float *fixed_vector, float voice_fac)
  746. {
  747.     int i;
  748.     float cpe  = 0.125 * (1 + voice_fac);
  749.     float last = fixed_vector[0]; // holds c(i - 1)
  750.  
  751.     fixed_vector[0] -= cpe * fixed_vector[1];
  752.  
  753.     for (i = 1; i < AMRWB_SFR_SIZE - 1; i++) {
  754.         float cur = fixed_vector[i];
  755.  
  756.         fixed_vector[i] -= cpe * (last + fixed_vector[i + 1]);
  757.         last = cur;
  758.     }
  759.  
  760.     fixed_vector[AMRWB_SFR_SIZE - 1] -= cpe * last;
  761. }
  762.  
  763. /**
  764.  * Conduct 16th order linear predictive coding synthesis from excitation.
  765.  *
  766.  * @param[in]     ctx              Pointer to the AMRWBContext
  767.  * @param[in]     lpc              Pointer to the LPC coefficients
  768.  * @param[out]    excitation       Buffer for synthesis final excitation
  769.  * @param[in]     fixed_gain       Fixed codebook gain for synthesis
  770.  * @param[in]     fixed_vector     Algebraic codebook vector
  771.  * @param[in,out] samples          Pointer to the output samples and memory
  772.  */
  773. static void synthesis(AMRWBContext *ctx, float *lpc, float *excitation,
  774.                       float fixed_gain, const float *fixed_vector,
  775.                       float *samples)
  776. {
  777.     ctx->acelpv_ctx.weighted_vector_sumf(excitation, ctx->pitch_vector, fixed_vector,
  778.                             ctx->pitch_gain[0], fixed_gain, AMRWB_SFR_SIZE);
  779.  
  780.     /* emphasize pitch vector contribution in low bitrate modes */
  781.     if (ctx->pitch_gain[0] > 0.5 && ctx->fr_cur_mode <= MODE_8k85) {
  782.         int i;
  783.         float energy = ctx->celpm_ctx.dot_productf(excitation, excitation,
  784.                                                     AMRWB_SFR_SIZE);
  785.  
  786.         // XXX: Weird part in both ref code and spec. A unknown parameter
  787.         // {beta} seems to be identical to the current pitch gain
  788.         float pitch_factor = 0.25 * ctx->pitch_gain[0] * ctx->pitch_gain[0];
  789.  
  790.         for (i = 0; i < AMRWB_SFR_SIZE; i++)
  791.             excitation[i] += pitch_factor * ctx->pitch_vector[i];
  792.  
  793.         ff_scale_vector_to_given_sum_of_squares(excitation, excitation,
  794.                                                 energy, AMRWB_SFR_SIZE);
  795.     }
  796.  
  797.     ctx->celpf_ctx.celp_lp_synthesis_filterf(samples, lpc, excitation,
  798.                                  AMRWB_SFR_SIZE, LP_ORDER);
  799. }
  800.  
  801. /**
  802.  * Apply to synthesis a de-emphasis filter of the form:
  803.  * H(z) = 1 / (1 - m * z^-1)
  804.  *
  805.  * @param[out]    out              Output buffer
  806.  * @param[in]     in               Input samples array with in[-1]
  807.  * @param[in]     m                Filter coefficient
  808.  * @param[in,out] mem              State from last filtering
  809.  */
  810. static void de_emphasis(float *out, float *in, float m, float mem[1])
  811. {
  812.     int i;
  813.  
  814.     out[0] = in[0] + m * mem[0];
  815.  
  816.     for (i = 1; i < AMRWB_SFR_SIZE; i++)
  817.          out[i] = in[i] + out[i - 1] * m;
  818.  
  819.     mem[0] = out[AMRWB_SFR_SIZE - 1];
  820. }
  821.  
  822. /**
  823.  * Upsample a signal by 5/4 ratio (from 12.8kHz to 16kHz) using
  824.  * a FIR interpolation filter. Uses past data from before *in address.
  825.  *
  826.  * @param[out] out                 Buffer for interpolated signal
  827.  * @param[in]  in                  Current signal data (length 0.8*o_size)
  828.  * @param[in]  o_size              Output signal length
  829.  * @param[in] ctx                  The context
  830.  */
  831. static void upsample_5_4(float *out, const float *in, int o_size, CELPMContext *ctx)
  832. {
  833.     const float *in0 = in - UPS_FIR_SIZE + 1;
  834.     int i, j, k;
  835.     int int_part = 0, frac_part;
  836.  
  837.     i = 0;
  838.     for (j = 0; j < o_size / 5; j++) {
  839.         out[i] = in[int_part];
  840.         frac_part = 4;
  841.         i++;
  842.  
  843.         for (k = 1; k < 5; k++) {
  844.             out[i] = ctx->dot_productf(in0 + int_part,
  845.                                                   upsample_fir[4 - frac_part],
  846.                                                   UPS_MEM_SIZE);
  847.             int_part++;
  848.             frac_part--;
  849.             i++;
  850.         }
  851.     }
  852. }
  853.  
  854. /**
  855.  * Calculate the high-band gain based on encoded index (23k85 mode) or
  856.  * on the low-band speech signal and the Voice Activity Detection flag.
  857.  *
  858.  * @param[in] ctx                  The context
  859.  * @param[in] synth                LB speech synthesis at 12.8k
  860.  * @param[in] hb_idx               Gain index for mode 23k85 only
  861.  * @param[in] vad                  VAD flag for the frame
  862.  */
  863. static float find_hb_gain(AMRWBContext *ctx, const float *synth,
  864.                           uint16_t hb_idx, uint8_t vad)
  865. {
  866.     int wsp = (vad > 0);
  867.     float tilt;
  868.  
  869.     if (ctx->fr_cur_mode == MODE_23k85)
  870.         return qua_hb_gain[hb_idx] * (1.0f / (1 << 14));
  871.  
  872.     tilt = ctx->celpm_ctx.dot_productf(synth, synth + 1, AMRWB_SFR_SIZE - 1) /
  873.            ctx->celpm_ctx.dot_productf(synth, synth, AMRWB_SFR_SIZE);
  874.  
  875.     /* return gain bounded by [0.1, 1.0] */
  876.     return av_clipf((1.0 - FFMAX(0.0, tilt)) * (1.25 - 0.25 * wsp), 0.1, 1.0);
  877. }
  878.  
  879. /**
  880.  * Generate the high-band excitation with the same energy from the lower
  881.  * one and scaled by the given gain.
  882.  *
  883.  * @param[in]  ctx                 The context
  884.  * @param[out] hb_exc              Buffer for the excitation
  885.  * @param[in]  synth_exc           Low-band excitation used for synthesis
  886.  * @param[in]  hb_gain             Wanted excitation gain
  887.  */
  888. static void scaled_hb_excitation(AMRWBContext *ctx, float *hb_exc,
  889.                                  const float *synth_exc, float hb_gain)
  890. {
  891.     int i;
  892.     float energy = ctx->celpm_ctx.dot_productf(synth_exc, synth_exc,
  893.                                                 AMRWB_SFR_SIZE);
  894.  
  895.     /* Generate a white-noise excitation */
  896.     for (i = 0; i < AMRWB_SFR_SIZE_16k; i++)
  897.         hb_exc[i] = 32768.0 - (uint16_t) av_lfg_get(&ctx->prng);
  898.  
  899.     ff_scale_vector_to_given_sum_of_squares(hb_exc, hb_exc,
  900.                                             energy * hb_gain * hb_gain,
  901.                                             AMRWB_SFR_SIZE_16k);
  902. }
  903.  
  904. /**
  905.  * Calculate the auto-correlation for the ISF difference vector.
  906.  */
  907. static float auto_correlation(float *diff_isf, float mean, int lag)
  908. {
  909.     int i;
  910.     float sum = 0.0;
  911.  
  912.     for (i = 7; i < LP_ORDER - 2; i++) {
  913.         float prod = (diff_isf[i] - mean) * (diff_isf[i - lag] - mean);
  914.         sum += prod * prod;
  915.     }
  916.     return sum;
  917. }
  918.  
  919. /**
  920.  * Extrapolate a ISF vector to the 16kHz range (20th order LP)
  921.  * used at mode 6k60 LP filter for the high frequency band.
  922.  *
  923.  * @param[out] isf Buffer for extrapolated isf; contains LP_ORDER
  924.  *                 values on input
  925.  */
  926. static void extrapolate_isf(float isf[LP_ORDER_16k])
  927. {
  928.     float diff_isf[LP_ORDER - 2], diff_mean;
  929.     float corr_lag[3];
  930.     float est, scale;
  931.     int i, j, i_max_corr;
  932.  
  933.     isf[LP_ORDER_16k - 1] = isf[LP_ORDER - 1];
  934.  
  935.     /* Calculate the difference vector */
  936.     for (i = 0; i < LP_ORDER - 2; i++)
  937.         diff_isf[i] = isf[i + 1] - isf[i];
  938.  
  939.     diff_mean = 0.0;
  940.     for (i = 2; i < LP_ORDER - 2; i++)
  941.         diff_mean += diff_isf[i] * (1.0f / (LP_ORDER - 4));
  942.  
  943.     /* Find which is the maximum autocorrelation */
  944.     i_max_corr = 0;
  945.     for (i = 0; i < 3; i++) {
  946.         corr_lag[i] = auto_correlation(diff_isf, diff_mean, i + 2);
  947.  
  948.         if (corr_lag[i] > corr_lag[i_max_corr])
  949.             i_max_corr = i;
  950.     }
  951.     i_max_corr++;
  952.  
  953.     for (i = LP_ORDER - 1; i < LP_ORDER_16k - 1; i++)
  954.         isf[i] = isf[i - 1] + isf[i - 1 - i_max_corr]
  955.                             - isf[i - 2 - i_max_corr];
  956.  
  957.     /* Calculate an estimate for ISF(18) and scale ISF based on the error */
  958.     est   = 7965 + (isf[2] - isf[3] - isf[4]) / 6.0;
  959.     scale = 0.5 * (FFMIN(est, 7600) - isf[LP_ORDER - 2]) /
  960.             (isf[LP_ORDER_16k - 2] - isf[LP_ORDER - 2]);
  961.  
  962.     for (i = LP_ORDER - 1, j = 0; i < LP_ORDER_16k - 1; i++, j++)
  963.         diff_isf[j] = scale * (isf[i] - isf[i - 1]);
  964.  
  965.     /* Stability insurance */
  966.     for (i = 1; i < LP_ORDER_16k - LP_ORDER; i++)
  967.         if (diff_isf[i] + diff_isf[i - 1] < 5.0) {
  968.             if (diff_isf[i] > diff_isf[i - 1]) {
  969.                 diff_isf[i - 1] = 5.0 - diff_isf[i];
  970.             } else
  971.                 diff_isf[i] = 5.0 - diff_isf[i - 1];
  972.         }
  973.  
  974.     for (i = LP_ORDER - 1, j = 0; i < LP_ORDER_16k - 1; i++, j++)
  975.         isf[i] = isf[i - 1] + diff_isf[j] * (1.0f / (1 << 15));
  976.  
  977.     /* Scale the ISF vector for 16000 Hz */
  978.     for (i = 0; i < LP_ORDER_16k - 1; i++)
  979.         isf[i] *= 0.8;
  980. }
  981.  
  982. /**
  983.  * Spectral expand the LP coefficients using the equation:
  984.  *   y[i] = x[i] * (gamma ** i)
  985.  *
  986.  * @param[out] out                 Output buffer (may use input array)
  987.  * @param[in]  lpc                 LP coefficients array
  988.  * @param[in]  gamma               Weighting factor
  989.  * @param[in]  size                LP array size
  990.  */
  991. static void lpc_weighting(float *out, const float *lpc, float gamma, int size)
  992. {
  993.     int i;
  994.     float fac = gamma;
  995.  
  996.     for (i = 0; i < size; i++) {
  997.         out[i] = lpc[i] * fac;
  998.         fac   *= gamma;
  999.     }
  1000. }
  1001.  
  1002. /**
  1003.  * Conduct 20th order linear predictive coding synthesis for the high
  1004.  * frequency band excitation at 16kHz.
  1005.  *
  1006.  * @param[in]     ctx              The context
  1007.  * @param[in]     subframe         Current subframe index (0 to 3)
  1008.  * @param[in,out] samples          Pointer to the output speech samples
  1009.  * @param[in]     exc              Generated white-noise scaled excitation
  1010.  * @param[in]     isf              Current frame isf vector
  1011.  * @param[in]     isf_past         Past frame final isf vector
  1012.  */
  1013. static void hb_synthesis(AMRWBContext *ctx, int subframe, float *samples,
  1014.                          const float *exc, const float *isf, const float *isf_past)
  1015. {
  1016.     float hb_lpc[LP_ORDER_16k];
  1017.     enum Mode mode = ctx->fr_cur_mode;
  1018.  
  1019.     if (mode == MODE_6k60) {
  1020.         float e_isf[LP_ORDER_16k]; // ISF vector for extrapolation
  1021.         double e_isp[LP_ORDER_16k];
  1022.  
  1023.         ctx->acelpv_ctx.weighted_vector_sumf(e_isf, isf_past, isf, isfp_inter[subframe],
  1024.                                 1.0 - isfp_inter[subframe], LP_ORDER);
  1025.  
  1026.         extrapolate_isf(e_isf);
  1027.  
  1028.         e_isf[LP_ORDER_16k - 1] *= 2.0;
  1029.         ff_acelp_lsf2lspd(e_isp, e_isf, LP_ORDER_16k);
  1030.         ff_amrwb_lsp2lpc(e_isp, hb_lpc, LP_ORDER_16k);
  1031.  
  1032.         lpc_weighting(hb_lpc, hb_lpc, 0.9, LP_ORDER_16k);
  1033.     } else {
  1034.         lpc_weighting(hb_lpc, ctx->lp_coef[subframe], 0.6, LP_ORDER);
  1035.     }
  1036.  
  1037.     ctx->celpf_ctx.celp_lp_synthesis_filterf(samples, hb_lpc, exc, AMRWB_SFR_SIZE_16k,
  1038.                                  (mode == MODE_6k60) ? LP_ORDER_16k : LP_ORDER);
  1039. }
  1040.  
  1041. /**
  1042.  * Apply a 15th order filter to high-band samples.
  1043.  * The filter characteristic depends on the given coefficients.
  1044.  *
  1045.  * @param[out]    out              Buffer for filtered output
  1046.  * @param[in]     fir_coef         Filter coefficients
  1047.  * @param[in,out] mem              State from last filtering (updated)
  1048.  * @param[in]     in               Input speech data (high-band)
  1049.  *
  1050.  * @remark It is safe to pass the same array in in and out parameters
  1051.  */
  1052.  
  1053. #ifndef hb_fir_filter
  1054. static void hb_fir_filter(float *out, const float fir_coef[HB_FIR_SIZE + 1],
  1055.                           float mem[HB_FIR_SIZE], const float *in)
  1056. {
  1057.     int i, j;
  1058.     float data[AMRWB_SFR_SIZE_16k + HB_FIR_SIZE]; // past and current samples
  1059.  
  1060.     memcpy(data, mem, HB_FIR_SIZE * sizeof(float));
  1061.     memcpy(data + HB_FIR_SIZE, in, AMRWB_SFR_SIZE_16k * sizeof(float));
  1062.  
  1063.     for (i = 0; i < AMRWB_SFR_SIZE_16k; i++) {
  1064.         out[i] = 0.0;
  1065.         for (j = 0; j <= HB_FIR_SIZE; j++)
  1066.             out[i] += data[i + j] * fir_coef[j];
  1067.     }
  1068.  
  1069.     memcpy(mem, data + AMRWB_SFR_SIZE_16k, HB_FIR_SIZE * sizeof(float));
  1070. }
  1071. #endif /* hb_fir_filter */
  1072.  
  1073. /**
  1074.  * Update context state before the next subframe.
  1075.  */
  1076. static void update_sub_state(AMRWBContext *ctx)
  1077. {
  1078.     memmove(&ctx->excitation_buf[0], &ctx->excitation_buf[AMRWB_SFR_SIZE],
  1079.             (AMRWB_P_DELAY_MAX + LP_ORDER + 1) * sizeof(float));
  1080.  
  1081.     memmove(&ctx->pitch_gain[1], &ctx->pitch_gain[0], 5 * sizeof(float));
  1082.     memmove(&ctx->fixed_gain[1], &ctx->fixed_gain[0], 1 * sizeof(float));
  1083.  
  1084.     memmove(&ctx->samples_az[0], &ctx->samples_az[AMRWB_SFR_SIZE],
  1085.             LP_ORDER * sizeof(float));
  1086.     memmove(&ctx->samples_up[0], &ctx->samples_up[AMRWB_SFR_SIZE],
  1087.             UPS_MEM_SIZE * sizeof(float));
  1088.     memmove(&ctx->samples_hb[0], &ctx->samples_hb[AMRWB_SFR_SIZE_16k],
  1089.             LP_ORDER_16k * sizeof(float));
  1090. }
  1091.  
  1092. static int amrwb_decode_frame(AVCodecContext *avctx, void *data,
  1093.                               int *got_frame_ptr, AVPacket *avpkt)
  1094. {
  1095.     AMRWBContext *ctx  = avctx->priv_data;
  1096.     AVFrame *frame     = data;
  1097.     AMRWBFrame   *cf   = &ctx->frame;
  1098.     const uint8_t *buf = avpkt->data;
  1099.     int buf_size       = avpkt->size;
  1100.     int expected_fr_size, header_size;
  1101.     float *buf_out;
  1102.     float spare_vector[AMRWB_SFR_SIZE];      // extra stack space to hold result from anti-sparseness processing
  1103.     float fixed_gain_factor;                 // fixed gain correction factor (gamma)
  1104.     float *synth_fixed_vector;               // pointer to the fixed vector that synthesis should use
  1105.     float synth_fixed_gain;                  // the fixed gain that synthesis should use
  1106.     float voice_fac, stab_fac;               // parameters used for gain smoothing
  1107.     float synth_exc[AMRWB_SFR_SIZE];         // post-processed excitation for synthesis
  1108.     float hb_exc[AMRWB_SFR_SIZE_16k];        // excitation for the high frequency band
  1109.     float hb_samples[AMRWB_SFR_SIZE_16k];    // filtered high-band samples from synthesis
  1110.     float hb_gain;
  1111.     int sub, i, ret;
  1112.  
  1113.     /* get output buffer */
  1114.     frame->nb_samples = 4 * AMRWB_SFR_SIZE_16k;
  1115.     if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
  1116.         return ret;
  1117.     buf_out = (float *)frame->data[0];
  1118.  
  1119.     header_size      = decode_mime_header(ctx, buf);
  1120.     if (ctx->fr_cur_mode > MODE_SID) {
  1121.         av_log(avctx, AV_LOG_ERROR,
  1122.                "Invalid mode %d\n", ctx->fr_cur_mode);
  1123.         return AVERROR_INVALIDDATA;
  1124.     }
  1125.     expected_fr_size = ((cf_sizes_wb[ctx->fr_cur_mode] + 7) >> 3) + 1;
  1126.  
  1127.     if (buf_size < expected_fr_size) {
  1128.         av_log(avctx, AV_LOG_ERROR,
  1129.             "Frame too small (%d bytes). Truncated file?\n", buf_size);
  1130.         *got_frame_ptr = 0;
  1131.         return AVERROR_INVALIDDATA;
  1132.     }
  1133.  
  1134.     if (!ctx->fr_quality || ctx->fr_cur_mode > MODE_SID)
  1135.         av_log(avctx, AV_LOG_ERROR, "Encountered a bad or corrupted frame\n");
  1136.  
  1137.     if (ctx->fr_cur_mode == MODE_SID) { /* Comfort noise frame */
  1138.         avpriv_request_sample(avctx, "SID mode");
  1139.         return AVERROR_PATCHWELCOME;
  1140.     }
  1141.  
  1142.     ff_amr_bit_reorder((uint16_t *) &ctx->frame, sizeof(AMRWBFrame),
  1143.         buf + header_size, amr_bit_orderings_by_mode[ctx->fr_cur_mode]);
  1144.  
  1145.     /* Decode the quantized ISF vector */
  1146.     if (ctx->fr_cur_mode == MODE_6k60) {
  1147.         decode_isf_indices_36b(cf->isp_id, ctx->isf_cur);
  1148.     } else {
  1149.         decode_isf_indices_46b(cf->isp_id, ctx->isf_cur);
  1150.     }
  1151.  
  1152.     isf_add_mean_and_past(ctx->isf_cur, ctx->isf_q_past);
  1153.     ff_set_min_dist_lsf(ctx->isf_cur, MIN_ISF_SPACING, LP_ORDER - 1);
  1154.  
  1155.     stab_fac = stability_factor(ctx->isf_cur, ctx->isf_past_final);
  1156.  
  1157.     ctx->isf_cur[LP_ORDER - 1] *= 2.0;
  1158.     ff_acelp_lsf2lspd(ctx->isp[3], ctx->isf_cur, LP_ORDER);
  1159.  
  1160.     /* Generate a ISP vector for each subframe */
  1161.     if (ctx->first_frame) {
  1162.         ctx->first_frame = 0;
  1163.         memcpy(ctx->isp_sub4_past, ctx->isp[3], LP_ORDER * sizeof(double));
  1164.     }
  1165.     interpolate_isp(ctx->isp, ctx->isp_sub4_past);
  1166.  
  1167.     for (sub = 0; sub < 4; sub++)
  1168.         ff_amrwb_lsp2lpc(ctx->isp[sub], ctx->lp_coef[sub], LP_ORDER);
  1169.  
  1170.     for (sub = 0; sub < 4; sub++) {
  1171.         const AMRWBSubFrame *cur_subframe = &cf->subframe[sub];
  1172.         float *sub_buf = buf_out + sub * AMRWB_SFR_SIZE_16k;
  1173.  
  1174.         /* Decode adaptive codebook (pitch vector) */
  1175.         decode_pitch_vector(ctx, cur_subframe, sub);
  1176.         /* Decode innovative codebook (fixed vector) */
  1177.         decode_fixed_vector(ctx->fixed_vector, cur_subframe->pul_ih,
  1178.                             cur_subframe->pul_il, ctx->fr_cur_mode);
  1179.  
  1180.         pitch_sharpening(ctx, ctx->fixed_vector);
  1181.  
  1182.         decode_gains(cur_subframe->vq_gain, ctx->fr_cur_mode,
  1183.                      &fixed_gain_factor, &ctx->pitch_gain[0]);
  1184.  
  1185.         ctx->fixed_gain[0] =
  1186.             ff_amr_set_fixed_gain(fixed_gain_factor,
  1187.                                   ctx->celpm_ctx.dot_productf(ctx->fixed_vector,
  1188.                                                                ctx->fixed_vector,
  1189.                                                                AMRWB_SFR_SIZE) /
  1190.                                   AMRWB_SFR_SIZE,
  1191.                        ctx->prediction_error,
  1192.                        ENERGY_MEAN, energy_pred_fac);
  1193.  
  1194.         /* Calculate voice factor and store tilt for next subframe */
  1195.         voice_fac      = voice_factor(ctx->pitch_vector, ctx->pitch_gain[0],
  1196.                                       ctx->fixed_vector, ctx->fixed_gain[0],
  1197.                                       &ctx->celpm_ctx);
  1198.         ctx->tilt_coef = voice_fac * 0.25 + 0.25;
  1199.  
  1200.         /* Construct current excitation */
  1201.         for (i = 0; i < AMRWB_SFR_SIZE; i++) {
  1202.             ctx->excitation[i] *= ctx->pitch_gain[0];
  1203.             ctx->excitation[i] += ctx->fixed_gain[0] * ctx->fixed_vector[i];
  1204.             ctx->excitation[i] = truncf(ctx->excitation[i]);
  1205.         }
  1206.  
  1207.         /* Post-processing of excitation elements */
  1208.         synth_fixed_gain = noise_enhancer(ctx->fixed_gain[0], &ctx->prev_tr_gain,
  1209.                                           voice_fac, stab_fac);
  1210.  
  1211.         synth_fixed_vector = anti_sparseness(ctx, ctx->fixed_vector,
  1212.                                              spare_vector);
  1213.  
  1214.         pitch_enhancer(synth_fixed_vector, voice_fac);
  1215.  
  1216.         synthesis(ctx, ctx->lp_coef[sub], synth_exc, synth_fixed_gain,
  1217.                   synth_fixed_vector, &ctx->samples_az[LP_ORDER]);
  1218.  
  1219.         /* Synthesis speech post-processing */
  1220.         de_emphasis(&ctx->samples_up[UPS_MEM_SIZE],
  1221.                     &ctx->samples_az[LP_ORDER], PREEMPH_FAC, ctx->demph_mem);
  1222.  
  1223.         ctx->acelpf_ctx.acelp_apply_order_2_transfer_function(&ctx->samples_up[UPS_MEM_SIZE],
  1224.             &ctx->samples_up[UPS_MEM_SIZE], hpf_zeros, hpf_31_poles,
  1225.             hpf_31_gain, ctx->hpf_31_mem, AMRWB_SFR_SIZE);
  1226.  
  1227.         upsample_5_4(sub_buf, &ctx->samples_up[UPS_FIR_SIZE],
  1228.                      AMRWB_SFR_SIZE_16k, &ctx->celpm_ctx);
  1229.  
  1230.         /* High frequency band (6.4 - 7.0 kHz) generation part */
  1231.         ctx->acelpf_ctx.acelp_apply_order_2_transfer_function(hb_samples,
  1232.             &ctx->samples_up[UPS_MEM_SIZE], hpf_zeros, hpf_400_poles,
  1233.             hpf_400_gain, ctx->hpf_400_mem, AMRWB_SFR_SIZE);
  1234.  
  1235.         hb_gain = find_hb_gain(ctx, hb_samples,
  1236.                                cur_subframe->hb_gain, cf->vad);
  1237.  
  1238.         scaled_hb_excitation(ctx, hb_exc, synth_exc, hb_gain);
  1239.  
  1240.         hb_synthesis(ctx, sub, &ctx->samples_hb[LP_ORDER_16k],
  1241.                      hb_exc, ctx->isf_cur, ctx->isf_past_final);
  1242.  
  1243.         /* High-band post-processing filters */
  1244.         hb_fir_filter(hb_samples, bpf_6_7_coef, ctx->bpf_6_7_mem,
  1245.                       &ctx->samples_hb[LP_ORDER_16k]);
  1246.  
  1247.         if (ctx->fr_cur_mode == MODE_23k85)
  1248.             hb_fir_filter(hb_samples, lpf_7_coef, ctx->lpf_7_mem,
  1249.                           hb_samples);
  1250.  
  1251.         /* Add the low and high frequency bands */
  1252.         for (i = 0; i < AMRWB_SFR_SIZE_16k; i++)
  1253.             sub_buf[i] = (sub_buf[i] + hb_samples[i]) * (1.0f / (1 << 15));
  1254.  
  1255.         /* Update buffers and history */
  1256.         update_sub_state(ctx);
  1257.     }
  1258.  
  1259.     /* update state for next frame */
  1260.     memcpy(ctx->isp_sub4_past, ctx->isp[3], LP_ORDER * sizeof(ctx->isp[3][0]));
  1261.     memcpy(ctx->isf_past_final, ctx->isf_cur, LP_ORDER * sizeof(float));
  1262.  
  1263.     *got_frame_ptr = 1;
  1264.  
  1265.     return expected_fr_size;
  1266. }
  1267.  
  1268. AVCodec ff_amrwb_decoder = {
  1269.     .name           = "amrwb",
  1270.     .long_name      = NULL_IF_CONFIG_SMALL("AMR-WB (Adaptive Multi-Rate WideBand)"),
  1271.     .type           = AVMEDIA_TYPE_AUDIO,
  1272.     .id             = AV_CODEC_ID_AMR_WB,
  1273.     .priv_data_size = sizeof(AMRWBContext),
  1274.     .init           = amrwb_decode_init,
  1275.     .decode         = amrwb_decode_frame,
  1276.     .capabilities   = CODEC_CAP_DR1,
  1277.     .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLT,
  1278.                                                      AV_SAMPLE_FMT_NONE },
  1279. };
  1280.