libavcodec/ra288.c

   1 /*
   2  * RealAudio 2.0 (28.8K)
   3  * Copyright (c) 2003 the ffmpeg project
   4  *
   5  * This file is part of Libav.
   6  *
   7  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #include "libavutil/channel_layout.h"
  23 #include "libavutil/float_dsp.h"
  24 #include "avcodec.h"
  25 #include "internal.h"
  26 #define BITSTREAM_READER_LE
  27 #include "get_bits.h"
  28 #include "ra288.h"
  29 #include "lpc.h"
  30 #include "celp_filters.h"
  31
  32 #define MAX_BACKWARD_FILTER_ORDER  36
  33 #define MAX_BACKWARD_FILTER_LEN    40
  34 #define MAX_BACKWARD_FILTER_NONREC 35
  35
  36 #define RA288_BLOCK_SIZE        5
  37 #define RA288_BLOCKS_PER_FRAME 32
  38
  39 typedef struct {
  40     AVFrame frame;
  41     DSPContext dsp;
  42     AVFloatDSPContext fdsp;
  43     DECLARE_ALIGNED(32, float,   sp_lpc)[FFALIGN(36, 16)];   ///< LPC coefficients for speech data (spec: A)
  44     DECLARE_ALIGNED(32, float, gain_lpc)[FFALIGN(10, 16)];   ///< LPC coefficients for gain        (spec: GB)
  45
  46     /** speech data history                                      (spec: SB).
  47      *  Its first 70 coefficients are updated only at backward filtering.
  48      */
  49     float sp_hist[111];
  50
  51     /// speech part of the gain autocorrelation                  (spec: REXP)
  52     float sp_rec[37];
  53
  54     /** log-gain history                                         (spec: SBLG).
  55      *  Its first 28 coefficients are updated only at backward filtering.
  56      */
  57     float gain_hist[38];
  58
  59     /// recursive part of the gain autocorrelation               (spec: REXPLG)
  60     float gain_rec[11];
  61 } RA288Context;
  62
  63 static av_cold int ra288_decode_init(AVCodecContext *avctx)
  64 {
  65     RA288Context *ractx = avctx->priv_data;
  66
  67     avctx->channels       = 1;
  68     avctx->channel_layout = AV_CH_LAYOUT_MONO;
  69     avctx->sample_fmt     = AV_SAMPLE_FMT_FLT;
  70
  71     avpriv_float_dsp_init(&ractx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
  72
  73     avcodec_get_frame_defaults(&ractx->frame);
  74     avctx->coded_frame = &ractx->frame;
  75
  76     return 0;
  77 }
  78
  79 static void convolve(float *tgt, const float *src, int len, int n)
  80 {
  81     for (; n >= 0; n--)
  82         tgt[n] = ff_scalarproduct_float_c(src, src - n, len);
  83
  84 }
  85
  86 static void decode(RA288Context *ractx, float gain, int cb_coef)
  87 {
  88     int i;
  89     double sumsum;
  90     float sum, buffer[5];
  91     float *block = ractx->sp_hist + 70 + 36; // current block
  92     float *gain_block = ractx->gain_hist + 28;
  93
  94     memmove(ractx->sp_hist + 70, ractx->sp_hist + 75, 36*sizeof(*block));
  95
  96     /* block 46 of G.728 spec */
  97     sum = 32.;
  98     for (i=0; i < 10; i++)
  99         sum -= gain_block[9-i] * ractx->gain_lpc[i];
 100
 101     /* block 47 of G.728 spec */
 102     sum = av_clipf(sum, 0, 60);
 103
 104     /* block 48 of G.728 spec */
 105     /* exp(sum * 0.1151292546497) == pow(10.0,sum/20) */
 106     sumsum = exp(sum * 0.1151292546497) * gain * (1.0/(1<<23));
 107
 108     for (i=0; i < 5; i++)
 109         buffer[i] = codetable[cb_coef][i] * sumsum;
 110
 111     sum = ff_scalarproduct_float_c(buffer, buffer, 5) * ((1 << 24) / 5.);
 112
 113     sum = FFMAX(sum, 1);
 114
 115     /* shift and store */
 116     memmove(gain_block, gain_block + 1, 9 * sizeof(*gain_block));
 117
 118     gain_block[9] = 10 * log10(sum) - 32;
 119
 120     ff_celp_lp_synthesis_filterf(block, ractx->sp_lpc, buffer, 5, 36);
 121 }
 122
 123 /**
 124  * Hybrid window filtering, see blocks 36 and 49 of the G.728 specification.
 125  *
 126  * @param order   filter order
 127  * @param n       input length
 128  * @param non_rec number of non-recursive samples
 129  * @param out     filter output
 130  * @param hist    pointer to the input history of the filter
 131  * @param out     pointer to the non-recursive part of the output
 132  * @param out2    pointer to the recursive part of the output
 133  * @param window  pointer to the windowing function table
 134  */
 135 static void do_hybrid_window(RA288Context *ractx,
 136                              int order, int n, int non_rec, float *out,
 137                              float *hist, float *out2, const float *window)
 138 {
 139     int i;
 140     float buffer1[MAX_BACKWARD_FILTER_ORDER + 1];
 141     float buffer2[MAX_BACKWARD_FILTER_ORDER + 1];
 142     LOCAL_ALIGNED(32, float, work, [FFALIGN(MAX_BACKWARD_FILTER_ORDER +
 143                                             MAX_BACKWARD_FILTER_LEN   +
 144                                             MAX_BACKWARD_FILTER_NONREC, 16)]);
 145
 146     ractx->fdsp.vector_fmul(work, window, hist, FFALIGN(order + n + non_rec, 16));
 147
 148     convolve(buffer1, work + order    , n      , order);
 149     convolve(buffer2, work + order + n, non_rec, order);
 150
 151     for (i=0; i <= order; i++) {
 152         out2[i] = out2[i] * 0.5625 + buffer1[i];
 153         out [i] = out2[i]          + buffer2[i];
 154     }
 155
 156     /* Multiply by the white noise correcting factor (WNCF). */
 157     *out *= 257./256.;
 158 }
 159
 160 /**
 161  * Backward synthesis filter, find the LPC coefficients from past speech data.
 162  */
 163 static void backward_filter(RA288Context *ractx,
 164                             float *hist, float *rec, const float *window,
 165                             float *lpc, const float *tab,
 166                             int order, int n, int non_rec, int move_size)
 167 {
 168     float temp[MAX_BACKWARD_FILTER_ORDER+1];
 169
 170     do_hybrid_window(ractx, order, n, non_rec, temp, hist, rec, window);
 171
 172     if (!compute_lpc_coefs(temp, order, lpc, 0, 1, 1))
 173         ractx->fdsp.vector_fmul(lpc, lpc, tab, FFALIGN(order, 16));
 174
 175     memmove(hist, hist + n, move_size*sizeof(*hist));
 176 }
 177
 178 static int ra288_decode_frame(AVCodecContext * avctx, void *data,
 179                               int *got_frame_ptr, AVPacket *avpkt)
 180 {
 181     const uint8_t *buf = avpkt->data;
 182     int buf_size = avpkt->size;
 183     float *out;
 184     int i, ret;
 185     RA288Context *ractx = avctx->priv_data;
 186     GetBitContext gb;
 187
 188     if (buf_size < avctx->block_align) {
 189         av_log(avctx, AV_LOG_ERROR,
 190                "Error! Input buffer is too small [%d<%d]\n",
 191                buf_size, avctx->block_align);
 192         return AVERROR_INVALIDDATA;
 193     }
 194
 195     /* get output buffer */
 196     ractx->frame.nb_samples = RA288_BLOCK_SIZE * RA288_BLOCKS_PER_FRAME;
 197     if ((ret = ff_get_buffer(avctx, &ractx->frame)) < 0) {
 198         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
 199         return ret;
 200     }
 201     out = (float *)ractx->frame.data[0];
 202
 203     init_get_bits(&gb, buf, avctx->block_align * 8);
 204
 205     for (i=0; i < RA288_BLOCKS_PER_FRAME; i++) {
 206         float gain = amptable[get_bits(&gb, 3)];
 207         int cb_coef = get_bits(&gb, 6 + (i&1));
 208
 209         decode(ractx, gain, cb_coef);
 210
 211         memcpy(out, &ractx->sp_hist[70 + 36], RA288_BLOCK_SIZE * sizeof(*out));
 212         out += RA288_BLOCK_SIZE;
 213
 214         if ((i & 7) == 3) {
 215             backward_filter(ractx, ractx->sp_hist, ractx->sp_rec, syn_window,
 216                             ractx->sp_lpc, syn_bw_tab, 36, 40, 35, 70);
 217
 218             backward_filter(ractx, ractx->gain_hist, ractx->gain_rec, gain_window,
 219                             ractx->gain_lpc, gain_bw_tab, 10, 8, 20, 28);
 220         }
 221     }
 222
 223     *got_frame_ptr   = 1;
 224     *(AVFrame *)data = ractx->frame;
 225
 226     return avctx->block_align;
 227 }
 228
 229 AVCodec ff_ra_288_decoder = {
 230     .name           = "real_288",
 231     .type           = AVMEDIA_TYPE_AUDIO,
 232     .id             = AV_CODEC_ID_RA_288,
 233     .priv_data_size = sizeof(RA288Context),
 234     .init           = ra288_decode_init,
 235     .decode         = ra288_decode_frame,
 236     .capabilities   = CODEC_CAP_DR1,
 237     .long_name      = NULL_IF_CONFIG_SMALL("RealAudio 2.0 (28.8K)"),
 238 };