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 "avcodec.h"
  23 #define ALT_BITSTREAM_READER_LE
  24 #include "bitstream.h"
  25 #include "ra288.h"
  26
  27 typedef struct {
  28     float sp_lpc[36];      ///< LPC coefficients for speech data (spec: A)
  29     float gain_lpc[10];    ///< LPC coefficients for gain (spec: GB)
  30
  31     float sp_hist[111];    ///< Speech data history (spec: SB)
  32
  33     /** Speech part of the gain autocorrelation (spec: REXP) */
  34     float sp_rec[37];
  35
  36     float gain_hist[38];   ///< Log-gain history (spec: SBLG)
  37
  38     /** Recursive part of the gain autocorrelation (spec: REXPLG) */
  39     float gain_rec[11];
  40
  41     float sp_block[41];    ///< Speech data of four blocks (spec: STTMP)
  42     float gain_block[10];  ///< Gain data of four blocks (spec: GSTATE)
  43 } RA288Context;
  44
  45 static av_cold int ra288_decode_init(AVCodecContext *avctx)
  46 {
  47     avctx->sample_fmt = SAMPLE_FMT_S16;
  48     return 0;
  49 }
  50
  51 static inline float scalar_product_float(const float * v1, const float * v2,
  52                                          int size)
  53 {
  54     float res = 0.;
  55
  56     while (size--)
  57         res += *v1++ * *v2++;
  58
  59     return res;
  60 }
  61
  62 static void colmult(float *tgt, const float *m1, const float *m2, int n)
  63 {
  64     while (n--)
  65         *tgt++ = *m1++ * *m2++;
  66 }
  67
  68 static void decode(RA288Context *ractx, float gain, int cb_coef)
  69 {
  70     int i, j;
  71     double sumsum;
  72     float sum, buffer[5];
  73
  74     memmove(ractx->sp_block + 5, ractx->sp_block, 36*sizeof(*ractx->sp_block));
  75
  76     for (i=4; i >= 0; i--)
  77         ractx->sp_block[i] = -scalar_product_float(ractx->sp_block + i + 1,
  78                                              ractx->sp_lpc, 36);
  79
  80     /* block 46 of G.728 spec */
  81     sum = 32. - scalar_product_float(ractx->gain_lpc, ractx->gain_block, 10);
  82
  83     /* block 47 of G.728 spec */
  84     sum = av_clipf(sum, 0, 60);
  85
  86     /* block 48 of G.728 spec */
  87     sumsum = exp(sum * 0.1151292546497) * gain; /* pow(10.0,sum/20)*gain */
  88
  89     for (i=0; i < 5; i++)
  90         buffer[i] = codetable[cb_coef][i] * sumsum;
  91
  92     sum = scalar_product_float(buffer, buffer, 5) / 5;
  93
  94     sum = FFMAX(sum, 1);
  95
  96     /* shift and store */
  97     memmove(ractx->gain_block, ractx->gain_block - 1,
  98             10 * sizeof(*ractx->gain_block));
  99
 100     *ractx->gain_block = 10 * log10(sum) - 32;
 101
 102     for (i=1; i < 5; i++)
 103         for (j=i-1; j >= 0; j--)
 104             buffer[i] -= ractx->sp_lpc[i-j-1] * buffer[j];
 105
 106     /* output */
 107     for (i=0; i < 5; i++)
 108         ractx->sp_block[4-i] =
 109             av_clipf(ractx->sp_block[4-i] + buffer[i], -4095, 4095);
 110 }
 111
 112 /**
 113  * Converts autocorrelation coefficients to LPC coefficients using the
 114  * Levinson-Durbin algorithm. See blocks 37 and 50 of the G.728 specification.
 115  *
 116  * @return 0 if success, -1 if fail
 117  */
 118 static int eval_lpc_coeffs(const float *in, float *tgt, int n)
 119 {
 120     int i, j;
 121     double f0, f1, f2;
 122
 123     if (in[n] == 0)
 124         return -1;
 125
 126     if ((f0 = *in) <= 0)
 127         return -1;
 128
 129     in--; // To avoid a -1 subtraction in the inner loop
 130
 131     for (i=1; i <= n; i++) {
 132         f1 = in[i+1];
 133
 134         for (j=0; j < i - 1; j++)
 135             f1 += in[i-j]*tgt[j];
 136
 137         tgt[i-1] = f2 = -f1/f0;
 138         for (j=0; j < i >> 1; j++) {
 139             float temp = tgt[j] + tgt[i-j-2]*f2;
 140             tgt[i-j-2] += tgt[j]*f2;
 141             tgt[j] = temp;
 142         }
 143         if ((f0 += f1*f2) < 0)
 144             return -1;
 145     }
 146
 147     return 0;
 148 }
 149
 150 static void convolve(float *tgt, const float *src, int len, int n)
 151 {
 152     for (; n >= 0; n--)
 153         tgt[n] = scalar_product_float(src, src - n, len);
 154
 155 }
 156
 157 /**
 158  * Hybrid window filtering. See blocks 36 and 49 of the G.728 specification.
 159  *
 160  * @note This function is slightly different from that described in the spec.
 161  *       It expects in[0] to be the newest sample and in[n-1] to be the oldest
 162  *       one stored. The spec has in the more ordinary way (in[0] the oldest
 163  *       and in[n-1] the newest).
 164  *
 165  * @param order   the order of the filter
 166  * @param n       the length of the input
 167  * @param non_rec the number of non-recursive samples
 168  * @param out     the filter output
 169  * @param in      pointer to the input of the filter
 170  * @param hist    pointer to the input history of the filter. It is updated by
 171  *                this function.
 172  * @param out     pointer to the non-recursive part of the output
 173  * @param out2    pointer to the recursive part of the output
 174  * @param window  pointer to the windowing function table
 175  */
 176 static void do_hybrid_window(int order, int n, int non_rec, const float *in,
 177                              float *out, float *hist, float *out2,
 178                              const float *window)
 179 {
 180     int i;
 181     float buffer1[order + 1];
 182     float buffer2[order + 1];
 183     float work[order + n + non_rec];
 184
 185     /* update history */
 186     memmove(hist, hist + n, (order + non_rec)*sizeof(*hist));
 187
 188     for (i=0; i < n; i++)
 189         hist[order + non_rec + i] = in[n-i-1];
 190
 191     colmult(work, window, hist, order + n + non_rec);
 192
 193     convolve(buffer1, work + order    , n      , order);
 194     convolve(buffer2, work + order + n, non_rec, order);
 195
 196     for (i=0; i <= order; i++) {
 197         out2[i] = out2[i] * 0.5625 + buffer1[i];
 198         out [i] = out2[i]          + buffer2[i];
 199     }
 200
 201     /* Multiply by the white noise correcting factor (WNCF) */
 202     *out *= 257./256.;
 203 }
 204
 205 /**
 206  * Backward synthesis filter. Find the LPC coefficients from past speech data.
 207  */
 208 static void backward_filter(RA288Context *ractx)
 209 {
 210     float temp1[37]; // RTMP in the spec
 211     float temp2[11]; // GPTPMP in the spec
 212
 213     do_hybrid_window(36, 40, 35, ractx->sp_block, temp1, ractx->sp_hist,
 214                      ractx->sp_rec, syn_window);
 215
 216     if (!eval_lpc_coeffs(temp1, ractx->sp_lpc, 36))
 217         colmult(ractx->sp_lpc, ractx->sp_lpc, syn_bw_tab, 36);
 218
 219     do_hybrid_window(10, 8, 20, ractx->gain_block, temp2, ractx->gain_hist,
 220                      ractx->gain_rec, gain_window);
 221
 222     if (!eval_lpc_coeffs(temp2, ractx->gain_lpc, 10))
 223         colmult(ractx->gain_lpc, ractx->gain_lpc, gain_bw_tab, 10);
 224 }
 225
 226 static int ra288_decode_frame(AVCodecContext * avctx, void *data,
 227                               int *data_size, const uint8_t * buf,
 228                               int buf_size)
 229 {
 230     int16_t *out = data;
 231     int i, j;
 232     RA288Context *ractx = avctx->priv_data;
 233     GetBitContext gb;
 234
 235     if (buf_size < avctx->block_align) {
 236         av_log(avctx, AV_LOG_ERROR,
 237                "Error! Input buffer is too small [%d<%d]\n",
 238                buf_size, avctx->block_align);
 239         return 0;
 240     }
 241
 242     init_get_bits(&gb, buf, avctx->block_align * 8);
 243
 244     for (i=0; i < 32; i++) {
 245         float gain = amptable[get_bits(&gb, 3)];
 246         int cb_coef = get_bits(&gb, 6 + (i&1));
 247
 248         decode(ractx, gain, cb_coef);
 249
 250         for (j=0; j < 5; j++)
 251             *(out++) = 8 * ractx->sp_block[4 - j];
 252
 253         if ((i & 7) == 3)
 254             backward_filter(ractx);
 255     }
 256
 257     *data_size = (char *)out - (char *)data;
 258     return avctx->block_align;
 259 }
 260
 261 AVCodec ra_288_decoder =
 262 {
 263     "real_288",
 264     CODEC_TYPE_AUDIO,
 265     CODEC_ID_RA_288,
 266     sizeof(RA288Context),
 267     ra288_decode_init,
 268     NULL,
 269     NULL,
 270     ra288_decode_frame,
 271     .long_name = NULL_IF_CONFIG_SMALL("RealAudio 2.0 (28.8K)"),
 272 };