ra288.c 8 KB
Newer Older
1 2
/*
 * RealAudio 2.0 (28.8K)
3
 * Copyright (c) 2003 The FFmpeg project
4
 *
5 6 7
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
8 9
 * modify it under the terms of the GNU Lesser General Public
 * 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 14 15 16 17
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * 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

26
#define BITSTREAM_READER_LE
27 28
#include "avcodec.h"
#include "celp_filters.h"
29
#include "get_bits.h"
30
#include "internal.h"
31
#include "lpc.h"
32
#include "ra288.h"
33

34 35 36 37
#define MAX_BACKWARD_FILTER_ORDER  36
#define MAX_BACKWARD_FILTER_LEN    40
#define MAX_BACKWARD_FILTER_NONREC 35

38 39 40
#define RA288_BLOCK_SIZE        5
#define RA288_BLOCKS_PER_FRAME 32

41
typedef struct RA288Context {
42
    AVFloatDSPContext *fdsp;
43 44
    DECLARE_ALIGNED(32, float,   sp_lpc)[FFALIGN(36, 16)];   ///< LPC coefficients for speech data (spec: A)
    DECLARE_ALIGNED(32, float, gain_lpc)[FFALIGN(10, 16)];   ///< LPC coefficients for gain        (spec: GB)
Vitor Sessak's avatar
Vitor Sessak committed
45

46 47 48 49
    /** speech data history                                      (spec: SB).
     *  Its first 70 coefficients are updated only at backward filtering.
     */
    float sp_hist[111];
50

51
    /// speech part of the gain autocorrelation                  (spec: REXP)
52 53
    float sp_rec[37];

54 55 56 57
    /** log-gain history                                         (spec: SBLG).
     *  Its first 28 coefficients are updated only at backward filtering.
     */
    float gain_hist[38];
58

59
    /// recursive part of the gain autocorrelation               (spec: REXPLG)
60
    float gain_rec[11];
61
} RA288Context;
62

63 64 65 66 67 68 69 70 71
static av_cold int ra288_decode_close(AVCodecContext *avctx)
{
    RA288Context *ractx = avctx->priv_data;

    av_freep(&ractx->fdsp);

    return 0;
}

72 73
static av_cold int ra288_decode_init(AVCodecContext *avctx)
{
74
    RA288Context *ractx = avctx->priv_data;
75 76 77 78 79

    avctx->channels       = 1;
    avctx->channel_layout = AV_CH_LAYOUT_MONO;
    avctx->sample_fmt     = AV_SAMPLE_FMT_FLT;

80
    if (avctx->block_align <= 0) {
81
        av_log(avctx, AV_LOG_ERROR, "unsupported block align\n");
82 83 84
        return AVERROR_PATCHWELCOME;
    }

85
    ractx->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
86 87
    if (!ractx->fdsp)
        return AVERROR(ENOMEM);
88

89 90 91
    return 0;
}

92 93 94
static void convolve(float *tgt, const float *src, int len, int n)
{
    for (; n >= 0; n--)
95
        tgt[n] = avpriv_scalarproduct_float_c(src, src - n, len);
96 97 98

}

99
static void decode(RA288Context *ractx, float gain, int cb_coef)
100
{
101
    int i;
102 103
    double sumsum;
    float sum, buffer[5];
104 105
    float *block = ractx->sp_hist + 70 + 36; // current block
    float *gain_block = ractx->gain_hist + 28;
106

107
    memmove(ractx->sp_hist + 70, ractx->sp_hist + 75, 36*sizeof(*block));
108

109
    /* block 46 of G.728 spec */
110
    sum = 32.0;
111
    for (i=0; i < 10; i++)
112
        sum -= gain_block[9-i] * ractx->gain_lpc[i];
113

114
    /* block 47 of G.728 spec */
Vitor Sessak's avatar
Vitor Sessak committed
115
    sum = av_clipf(sum, 0, 60);
Vitor Sessak's avatar
Vitor Sessak committed
116

117
    /* block 48 of G.728 spec */
118
    /* exp(sum * 0.1151292546497) == pow(10.0,sum/20) */
119
    sumsum = exp(sum * 0.1151292546497) * gain * (1.0/(1<<23));
Vitor Sessak's avatar
Vitor Sessak committed
120

121
    for (i=0; i < 5; i++)
122
        buffer[i] = codetable[cb_coef][i] * sumsum;
Vitor Sessak's avatar
Vitor Sessak committed
123

124
    sum = avpriv_scalarproduct_float_c(buffer, buffer, 5);
125

126
    sum = FFMAX(sum, 5.0 / (1<<24));
Vitor Sessak's avatar
Vitor Sessak committed
127 128

    /* shift and store */
129
    memmove(gain_block, gain_block + 1, 9 * sizeof(*gain_block));
Vitor Sessak's avatar
Vitor Sessak committed
130

131
    gain_block[9] = 10 * log10(sum) + (10*log10(((1<<24)/5.)) - 32);
Vitor Sessak's avatar
Vitor Sessak committed
132

133
    ff_celp_lp_synthesis_filterf(block, ractx->sp_lpc, buffer, 5, 36);
134 135
}

136
/**
137
 * Hybrid window filtering, see blocks 36 and 49 of the G.728 specification.
138
 *
139 140 141 142
 * @param order   filter order
 * @param n       input length
 * @param non_rec number of non-recursive samples
 * @param out     filter output
143
 * @param hist    pointer to the input history of the filter
Vitor Sessak's avatar
Vitor Sessak committed
144
 * @param out     pointer to the non-recursive part of the output
145 146 147
 * @param out2    pointer to the recursive part of the output
 * @param window  pointer to the windowing function table
 */
148 149
static void do_hybrid_window(RA288Context *ractx,
                             int order, int n, int non_rec, float *out,
Vitor Sessak's avatar
Vitor Sessak committed
150
                             float *hist, float *out2, const float *window)
151
{
152
    int i;
153 154
    float buffer1[MAX_BACKWARD_FILTER_ORDER + 1];
    float buffer2[MAX_BACKWARD_FILTER_ORDER + 1];
155 156 157
    LOCAL_ALIGNED(32, float, work, [FFALIGN(MAX_BACKWARD_FILTER_ORDER +
                                            MAX_BACKWARD_FILTER_LEN   +
                                            MAX_BACKWARD_FILTER_NONREC, 16)]);
Vitor Sessak's avatar
Vitor Sessak committed
158

159 160
    av_assert2(order>=0);

161
    ractx->fdsp->vector_fmul(work, window, hist, FFALIGN(order + n + non_rec, 16));
162

163 164
    convolve(buffer1, work + order    , n      , order);
    convolve(buffer2, work + order + n, non_rec, order);
165

166 167 168
    for (i=0; i <= order; i++) {
        out2[i] = out2[i] * 0.5625 + buffer1[i];
        out [i] = out2[i]          + buffer2[i];
Vitor Sessak's avatar
Vitor Sessak committed
169
    }
170

171
    /* Multiply by the white noise correcting factor (WNCF). */
172
    *out *= 257.0 / 256.0;
173 174
}

175
/**
176
 * Backward synthesis filter, find the LPC coefficients from past speech data.
177
 */
178 179
static void backward_filter(RA288Context *ractx,
                            float *hist, float *rec, const float *window,
180 181
                            float *lpc, const float *tab,
                            int order, int n, int non_rec, int move_size)
182
{
183
    float temp[MAX_BACKWARD_FILTER_ORDER+1];
184

185
    do_hybrid_window(ractx, order, n, non_rec, temp, hist, rec, window);
186

187
    if (!compute_lpc_coefs(temp, order, lpc, 0, 1, 1))
188
        ractx->fdsp->vector_fmul(lpc, lpc, tab, FFALIGN(order, 16));
189

190
    memmove(hist, hist + n, move_size*sizeof(*hist));
191 192
}

193
static int ra288_decode_frame(AVCodecContext * avctx, void *data,
194
                              int *got_frame_ptr, AVPacket *avpkt)
Nick Kurshev's avatar
Nick Kurshev committed
195
{
196
    AVFrame *frame     = data;
197 198
    const uint8_t *buf = avpkt->data;
    int buf_size = avpkt->size;
199 200
    float *out;
    int i, ret;
201
    RA288Context *ractx = avctx->priv_data;
202
    GetBitContext gb;
Vitor Sessak's avatar
Vitor Sessak committed
203

204 205 206 207
    if (buf_size < avctx->block_align) {
        av_log(avctx, AV_LOG_ERROR,
               "Error! Input buffer is too small [%d<%d]\n",
               buf_size, avctx->block_align);
208
        return AVERROR_INVALIDDATA;
209 210
    }

211 212 213 214
    ret = init_get_bits8(&gb, buf, avctx->block_align);
    if (ret < 0)
        return ret;

215
    /* get output buffer */
216
    frame->nb_samples = RA288_BLOCK_SIZE * RA288_BLOCKS_PER_FRAME;
217
    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
218
        return ret;
219
    out = (float *)frame->data[0];
220

221
    for (i=0; i < RA288_BLOCKS_PER_FRAME; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
222
        float gain = amptable[get_bits(&gb, 3)];
223
        int cb_coef = get_bits(&gb, 6 + (i&1));
Vitor Sessak's avatar
Vitor Sessak committed
224

225
        decode(ractx, gain, cb_coef);
Vitor Sessak's avatar
Vitor Sessak committed
226

227 228
        memcpy(out, &ractx->sp_hist[70 + 36], RA288_BLOCK_SIZE * sizeof(*out));
        out += RA288_BLOCK_SIZE;
Vitor Sessak's avatar
Vitor Sessak committed
229

230
        if ((i & 7) == 3) {
231
            backward_filter(ractx, ractx->sp_hist, ractx->sp_rec, syn_window,
232 233
                            ractx->sp_lpc, syn_bw_tab, 36, 40, 35, 70);

234
            backward_filter(ractx, ractx->gain_hist, ractx->gain_rec, gain_window,
235 236
                            ractx->gain_lpc, gain_bw_tab, 10, 8, 20, 28);
        }
Vitor Sessak's avatar
Vitor Sessak committed
237 238
    }

239
    *got_frame_ptr = 1;
240

241
    return avctx->block_align;
242 243
}

244 245
AVCodec ff_ra_288_decoder = {
    .name           = "real_288",
246
    .long_name      = NULL_IF_CONFIG_SMALL("RealAudio 2.0 (28.8K)"),
247
    .type           = AVMEDIA_TYPE_AUDIO,
248
    .id             = AV_CODEC_ID_RA_288,
249 250 251
    .priv_data_size = sizeof(RA288Context),
    .init           = ra288_decode_init,
    .decode         = ra288_decode_frame,
252
    .close          = ra288_decode_close,
253
    .capabilities   = AV_CODEC_CAP_DR1,
254
};