ffmpeg/libavcodec/ra288.c

   1 /*
   2  * RealAudio 2.0 (28.8K)
   3  * Copyright (c) 2003 The FFmpeg Project
   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 #include "libavutil/channel_layout.h"
  23 #include "libavutil/float_dsp.h"
  24 #include "libavutil/internal.h"
  25 #include "avcodec.h"
  26 #include "internal.h"
  27 #define BITSTREAM_READER_LE
  28 #include "get_bits.h"
  29 #include "ra288.h"
  30 #include "lpc.h"
  31 #include "celp_filters.h"
  32
  33 #define MAX_BACKWARD_FILTER_ORDER  36
  34 #define MAX_BACKWARD_FILTER_LEN    40
  35 #define MAX_BACKWARD_FILTER_NONREC 35
  36
  37 #define RA288_BLOCK_SIZE        5
  38 #define RA288_BLOCKS_PER_FRAME 32
  39
  40 typedef struct {
  41     AVFloatDSPContext *fdsp;
  42     DECLARE_ALIGNED(32, float,   sp_lpc)[FFALIGN(36, 16)];   ///< LPC coefficients for speech data (spec: A)
  43     DECLARE_ALIGNED(32, float, gain_lpc)[FFALIGN(10, 16)];   ///< LPC coefficients for gain        (spec: GB)
  44
  45     /** speech data history                                      (spec: SB).
  46      *  Its first 70 coefficients are updated only at backward filtering.
  47      */
  48     float sp_hist[111];
  49
  50     /// speech part of the gain autocorrelation                  (spec: REXP)
  51     float sp_rec[37];
  52
  53     /** log-gain history                                         (spec: SBLG).
  54      *  Its first 28 coefficients are updated only at backward filtering.
  55      */
  56     float gain_hist[38];
  57
  58     /// recursive part of the gain autocorrelation               (spec: REXPLG)
  59     float gain_rec[11];
  60 } RA288Context;
  61
  62 static av_cold int ra288_decode_close(AVCodecContext *avctx)
  63 {
  64     RA288Context *ractx = avctx->priv_data;
  65
  66     av_freep(&ractx->fdsp);
  67
  68     return 0;
  69 }
  70
  71 static av_cold int ra288_decode_init(AVCodecContext *avctx)
  72 {
  73     RA288Context *ractx = avctx->priv_data;
  74
  75     avctx->channels       = 1;
  76     avctx->channel_layout = AV_CH_LAYOUT_MONO;
  77     avctx->sample_fmt     = AV_SAMPLE_FMT_FLT;
  78
  79     if (avctx->block_align <= 0) {
  80         av_log(avctx, AV_LOG_ERROR, "unsupported block align\n");
  81         return AVERROR_PATCHWELCOME;
  82     }
  83
  84     ractx->fdsp = avpriv_float_dsp_alloc(avctx->flags & CODEC_FLAG_BITEXACT);
  85     if (!ractx->fdsp)
  86         return AVERROR(ENOMEM);
  87
  88     return 0;
  89 }
  90
  91 static void convolve(float *tgt, const float *src, int len, int n)
  92 {
  93     for (; n >= 0; n--)
  94         tgt[n] = avpriv_scalarproduct_float_c(src, src - n, len);
  95
  96 }
  97
  98 static void decode(RA288Context *ractx, float gain, int cb_coef)
  99 {
 100     int i;
 101     double sumsum;
 102     float sum, buffer[5];
 103     float *block = ractx->sp_hist + 70 + 36; // current block
 104     float *gain_block = ractx->gain_hist + 28;
 105
 106     memmove(ractx->sp_hist + 70, ractx->sp_hist + 75, 36*sizeof(*block));
 107
 108     /* block 46 of G.728 spec */
 109     sum = 32.0;
 110     for (i=0; i < 10; i++)
 111         sum -= gain_block[9-i] * ractx->gain_lpc[i];
 112
 113     /* block 47 of G.728 spec */
 114     sum = av_clipf(sum, 0, 60);
 115
 116     /* block 48 of G.728 spec */
 117     /* exp(sum * 0.1151292546497) == pow(10.0,sum/20) */
 118     sumsum = exp(sum * 0.1151292546497) * gain * (1.0/(1<<23));
 119
 120     for (i=0; i < 5; i++)
 121         buffer[i] = codetable[cb_coef][i] * sumsum;
 122
 123     sum = avpriv_scalarproduct_float_c(buffer, buffer, 5);
 124
 125     sum = FFMAX(sum, 5.0 / (1<<24));
 126
 127     /* shift and store */
 128     memmove(gain_block, gain_block + 1, 9 * sizeof(*gain_block));
 129
 130     gain_block[9] = 10 * log10(sum) + (10*log10(((1<<24)/5.)) - 32);
 131
 132     ff_celp_lp_synthesis_filterf(block, ractx->sp_lpc, buffer, 5, 36);
 133 }
 134
 135 /**
 136  * Hybrid window filtering, see blocks 36 and 49 of the G.728 specification.
 137  *
 138  * @param order   filter order
 139  * @param n       input length
 140  * @param non_rec number of non-recursive samples
 141  * @param out     filter output
 142  * @param hist    pointer to the input history of the filter
 143  * @param out     pointer to the non-recursive part of the output
 144  * @param out2    pointer to the recursive part of the output
 145  * @param window  pointer to the windowing function table
 146  */
 147 static void do_hybrid_window(RA288Context *ractx,
 148                              int order, int n, int non_rec, float *out,
 149                              float *hist, float *out2, const float *window)
 150 {
 151     int i;
 152     float buffer1[MAX_BACKWARD_FILTER_ORDER + 1];
 153     float buffer2[MAX_BACKWARD_FILTER_ORDER + 1];
 154     LOCAL_ALIGNED(32, float, work, [FFALIGN(MAX_BACKWARD_FILTER_ORDER +
 155                                             MAX_BACKWARD_FILTER_LEN   +
 156                                             MAX_BACKWARD_FILTER_NONREC, 16)]);
 157
 158     av_assert2(order>=0);
 159
 160     ractx->fdsp->vector_fmul(work, window, hist, FFALIGN(order + n + non_rec, 16));
 161
 162     convolve(buffer1, work + order    , n      , order);
 163     convolve(buffer2, work + order + n, non_rec, order);
 164
 165     for (i=0; i <= order; i++) {
 166         out2[i] = out2[i] * 0.5625 + buffer1[i];
 167         out [i] = out2[i]          + buffer2[i];
 168     }
 169
 170     /* Multiply by the white noise correcting factor (WNCF). */
 171     *out *= 257.0 / 256.0;
 172 }
 173
 174 /**
 175  * Backward synthesis filter, find the LPC coefficients from past speech data.
 176  */
 177 static void backward_filter(RA288Context *ractx,
 178                             float *hist, float *rec, const float *window,
 179                             float *lpc, const float *tab,
 180                             int order, int n, int non_rec, int move_size)
 181 {
 182     float temp[MAX_BACKWARD_FILTER_ORDER+1];
 183
 184     do_hybrid_window(ractx, order, n, non_rec, temp, hist, rec, window);
 185
 186     if (!compute_lpc_coefs(temp, order, lpc, 0, 1, 1))
 187         ractx->fdsp->vector_fmul(lpc, lpc, tab, FFALIGN(order, 16));
 188
 189     memmove(hist, hist + n, move_size*sizeof(*hist));
 190 }
 191
 192 static int ra288_decode_frame(AVCodecContext * avctx, void *data,
 193                               int *got_frame_ptr, AVPacket *avpkt)
 194 {
 195     AVFrame *frame     = data;
 196     const uint8_t *buf = avpkt->data;
 197     int buf_size = avpkt->size;
 198     float *out;
 199     int i, ret;
 200     RA288Context *ractx = avctx->priv_data;
 201     GetBitContext gb;
 202
 203     if (buf_size < avctx->block_align) {
 204         av_log(avctx, AV_LOG_ERROR,
 205                "Error! Input buffer is too small [%d<%d]\n",
 206                buf_size, avctx->block_align);
 207         return AVERROR_INVALIDDATA;
 208     }
 209
 210     /* get output buffer */
 211     frame->nb_samples = RA288_BLOCK_SIZE * RA288_BLOCKS_PER_FRAME;
 212     if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
 213         return ret;
 214     out = (float *)frame->data[0];
 215
 216     init_get_bits8(&gb, buf, avctx->block_align);
 217
 218     for (i=0; i < RA288_BLOCKS_PER_FRAME; i++) {
 219         float gain = amptable[get_bits(&gb, 3)];
 220         int cb_coef = get_bits(&gb, 6 + (i&1));
 221
 222         decode(ractx, gain, cb_coef);
 223
 224         memcpy(out, &ractx->sp_hist[70 + 36], RA288_BLOCK_SIZE * sizeof(*out));
 225         out += RA288_BLOCK_SIZE;
 226
 227         if ((i & 7) == 3) {
 228             backward_filter(ractx, ractx->sp_hist, ractx->sp_rec, syn_window,
 229                             ractx->sp_lpc, syn_bw_tab, 36, 40, 35, 70);
 230
 231             backward_filter(ractx, ractx->gain_hist, ractx->gain_rec, gain_window,
 232                             ractx->gain_lpc, gain_bw_tab, 10, 8, 20, 28);
 233         }
 234     }
 235
 236     *got_frame_ptr = 1;
 237
 238     return avctx->block_align;
 239 }
 240
 241 AVCodec ff_ra_288_decoder = {
 242     .name           = "real_288",
 243     .long_name      = NULL_IF_CONFIG_SMALL("RealAudio 2.0 (28.8K)"),
 244     .type           = AVMEDIA_TYPE_AUDIO,
 245     .id             = AV_CODEC_ID_RA_288,
 246     .priv_data_size = sizeof(RA288Context),
 247     .init           = ra288_decode_init,
 248     .decode         = ra288_decode_frame,
 249     .close          = ra288_decode_close,
 250     .capabilities   = CODEC_CAP_DR1,
 251 };