ra288.c 7.86 KB
Newer Older
1 2 3 4
/*
 * RealAudio 2.0 (28.8K)
 * Copyright (c) 2003 the ffmpeg project
 *
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 "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 33 34 35
#define MAX_BACKWARD_FILTER_ORDER  36
#define MAX_BACKWARD_FILTER_LEN    40
#define MAX_BACKWARD_FILTER_NONREC 35

36 37 38
#define RA288_BLOCK_SIZE        5
#define RA288_BLOCKS_PER_FRAME 32

39
typedef struct {
40
    AVFrame frame;
41
    DSPContext dsp;
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
static av_cold int ra288_decode_init(AVCodecContext *avctx)
{
65
    RA288Context *ractx = avctx->priv_data;
66 67 68 69 70

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

71 72 73 74 75
    if (avctx->block_align <= 0) {
        av_log_ask_for_sample(avctx, "unsupported block align\n");
        return AVERROR_PATCHWELCOME;
    }

76
    avpriv_float_dsp_init(&ractx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
77 78 79 80

    avcodec_get_frame_defaults(&ractx->frame);
    avctx->coded_frame = &ractx->frame;

81 82 83
    return 0;
}

84 85 86
static void convolve(float *tgt, const float *src, int len, int n)
{
    for (; n >= 0; n--)
87
        tgt[n] = ff_scalarproduct_float_c(src, src - n, len);
88 89 90

}

91
static void decode(RA288Context *ractx, float gain, int cb_coef)
92
{
93
    int i;
94 95
    double sumsum;
    float sum, buffer[5];
96 97
    float *block = ractx->sp_hist + 70 + 36; // current block
    float *gain_block = ractx->gain_hist + 28;
98

99
    memmove(ractx->sp_hist + 70, ractx->sp_hist + 75, 36*sizeof(*block));
100

101
    /* block 46 of G.728 spec */
102 103
    sum = 32.;
    for (i=0; i < 10; i++)
104
        sum -= gain_block[9-i] * ractx->gain_lpc[i];
105

106
    /* block 47 of G.728 spec */
Vitor Sessak's avatar
Vitor Sessak committed
107
    sum = av_clipf(sum, 0, 60);
Vitor Sessak's avatar
Vitor Sessak committed
108

109
    /* block 48 of G.728 spec */
110
    /* exp(sum * 0.1151292546497) == pow(10.0,sum/20) */
111
    sumsum = exp(sum * 0.1151292546497) * gain * (1.0/(1<<23));
Vitor Sessak's avatar
Vitor Sessak committed
112

113
    for (i=0; i < 5; i++)
114
        buffer[i] = codetable[cb_coef][i] * sumsum;
Vitor Sessak's avatar
Vitor Sessak committed
115

116
    sum = ff_scalarproduct_float_c(buffer, buffer, 5);
117

118
    sum = FFMAX(sum, 5. / (1<<24));
Vitor Sessak's avatar
Vitor Sessak committed
119 120

    /* shift and store */
121
    memmove(gain_block, gain_block + 1, 9 * sizeof(*gain_block));
Vitor Sessak's avatar
Vitor Sessak committed
122

123
    gain_block[9] = 10 * log10(sum) + (10*log10(((1<<24)/5.)) - 32);
Vitor Sessak's avatar
Vitor Sessak committed
124

125
    ff_celp_lp_synthesis_filterf(block, ractx->sp_lpc, buffer, 5, 36);
126 127
}

128
/**
129
 * Hybrid window filtering, see blocks 36 and 49 of the G.728 specification.
130
 *
131 132 133 134
 * @param order   filter order
 * @param n       input length
 * @param non_rec number of non-recursive samples
 * @param out     filter output
135
 * @param hist    pointer to the input history of the filter
Vitor Sessak's avatar
Vitor Sessak committed
136
 * @param out     pointer to the non-recursive part of the output
137 138 139
 * @param out2    pointer to the recursive part of the output
 * @param window  pointer to the windowing function table
 */
140 141
static void do_hybrid_window(RA288Context *ractx,
                             int order, int n, int non_rec, float *out,
Vitor Sessak's avatar
Vitor Sessak committed
142
                             float *hist, float *out2, const float *window)
143
{
144
    int i;
145 146
    float buffer1[MAX_BACKWARD_FILTER_ORDER + 1];
    float buffer2[MAX_BACKWARD_FILTER_ORDER + 1];
147 148 149
    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
150

151 152
    av_assert2(order>=0);

153
    ractx->fdsp.vector_fmul(work, window, hist, FFALIGN(order + n + non_rec, 16));
154

155 156
    convolve(buffer1, work + order    , n      , order);
    convolve(buffer2, work + order + n, non_rec, order);
157

158 159 160
    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
161
    }
162

163
    /* Multiply by the white noise correcting factor (WNCF). */
164
    *out *= 257./256.;
165 166
}

167
/**
168
 * Backward synthesis filter, find the LPC coefficients from past speech data.
169
 */
170 171
static void backward_filter(RA288Context *ractx,
                            float *hist, float *rec, const float *window,
172 173
                            float *lpc, const float *tab,
                            int order, int n, int non_rec, int move_size)
174
{
175
    float temp[MAX_BACKWARD_FILTER_ORDER+1];
176

177
    do_hybrid_window(ractx, order, n, non_rec, temp, hist, rec, window);
178

179
    if (!compute_lpc_coefs(temp, order, lpc, 0, 1, 1))
180
        ractx->fdsp.vector_fmul(lpc, lpc, tab, FFALIGN(order, 16));
181

182
    memmove(hist, hist + n, move_size*sizeof(*hist));
183 184
}

185
static int ra288_decode_frame(AVCodecContext * avctx, void *data,
186
                              int *got_frame_ptr, AVPacket *avpkt)
Nick Kurshev's avatar
Nick Kurshev committed
187
{
188 189
    const uint8_t *buf = avpkt->data;
    int buf_size = avpkt->size;
190 191
    float *out;
    int i, ret;
192
    RA288Context *ractx = avctx->priv_data;
193
    GetBitContext gb;
Vitor Sessak's avatar
Vitor Sessak committed
194

195 196 197 198
    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);
199
        return AVERROR_INVALIDDATA;
200 201
    }

202 203
    /* get output buffer */
    ractx->frame.nb_samples = RA288_BLOCK_SIZE * RA288_BLOCKS_PER_FRAME;
204
    if ((ret = ff_get_buffer(avctx, &ractx->frame)) < 0) {
205 206
        av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
        return ret;
207
    }
208
    out = (float *)ractx->frame.data[0];
209

210
    init_get_bits(&gb, buf, avctx->block_align * 8);
Vitor Sessak's avatar
Vitor Sessak committed
211

212
    for (i=0; i < RA288_BLOCKS_PER_FRAME; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
213
        float gain = amptable[get_bits(&gb, 3)];
214
        int cb_coef = get_bits(&gb, 6 + (i&1));
Vitor Sessak's avatar
Vitor Sessak committed
215

216
        decode(ractx, gain, cb_coef);
Vitor Sessak's avatar
Vitor Sessak committed
217

218 219
        memcpy(out, &ractx->sp_hist[70 + 36], RA288_BLOCK_SIZE * sizeof(*out));
        out += RA288_BLOCK_SIZE;
Vitor Sessak's avatar
Vitor Sessak committed
220

221
        if ((i & 7) == 3) {
222
            backward_filter(ractx, ractx->sp_hist, ractx->sp_rec, syn_window,
223 224
                            ractx->sp_lpc, syn_bw_tab, 36, 40, 35, 70);

225
            backward_filter(ractx, ractx->gain_hist, ractx->gain_rec, gain_window,
226 227
                            ractx->gain_lpc, gain_bw_tab, 10, 8, 20, 28);
        }
Vitor Sessak's avatar
Vitor Sessak committed
228 229
    }

230 231 232
    *got_frame_ptr   = 1;
    *(AVFrame *)data = ractx->frame;

233
    return avctx->block_align;
234 235
}

236 237 238
AVCodec ff_ra_288_decoder = {
    .name           = "real_288",
    .type           = AVMEDIA_TYPE_AUDIO,
239
    .id             = AV_CODEC_ID_RA_288,
240 241 242
    .priv_data_size = sizeof(RA288Context),
    .init           = ra288_decode_init,
    .decode         = ra288_decode_frame,
243
    .capabilities   = CODEC_CAP_DR1,
244
    .long_name      = NULL_IF_CONFIG_SMALL("RealAudio 2.0 (28.8K)"),
245
};