ra288.c 7.95 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
#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 34 35 36
#define MAX_BACKWARD_FILTER_ORDER  36
#define MAX_BACKWARD_FILTER_LEN    40
#define MAX_BACKWARD_FILTER_NONREC 35

37 38 39
#define RA288_BLOCK_SIZE        5
#define RA288_BLOCKS_PER_FRAME 32

40
typedef struct RA288Context {
41
    AVFloatDSPContext *fdsp;
42 43
    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
44

45 46 47 48
    /** speech data history                                      (spec: SB).
     *  Its first 70 coefficients are updated only at backward filtering.
     */
    float sp_hist[111];
49

50
    /// speech part of the gain autocorrelation                  (spec: REXP)
51 52
    float sp_rec[37];

53 54 55 56
    /** log-gain history                                         (spec: SBLG).
     *  Its first 28 coefficients are updated only at backward filtering.
     */
    float gain_hist[38];
57

58
    /// recursive part of the gain autocorrelation               (spec: REXPLG)
59
    float gain_rec[11];
60
} RA288Context;
61

62 63 64 65 66 67 68 69 70
static av_cold int ra288_decode_close(AVCodecContext *avctx)
{
    RA288Context *ractx = avctx->priv_data;

    av_freep(&ractx->fdsp);

    return 0;
}

71 72
static av_cold int ra288_decode_init(AVCodecContext *avctx)
{
73
    RA288Context *ractx = avctx->priv_data;
74 75 76 77 78

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

79
    if (avctx->block_align <= 0) {
80
        av_log(avctx, AV_LOG_ERROR, "unsupported block align\n");
81 82 83
        return AVERROR_PATCHWELCOME;
    }

84 85 86
    ractx->fdsp = avpriv_float_dsp_alloc(avctx->flags & CODEC_FLAG_BITEXACT);
    if (!ractx->fdsp)
        return AVERROR(ENOMEM);
87

88 89 90
    return 0;
}

91 92 93
static void convolve(float *tgt, const float *src, int len, int n)
{
    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;
101 102
    double sumsum;
    float sum, buffer[5];
103 104
    float *block = ractx->sp_hist + 70 + 36; // current block
    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 */
Vitor Sessak's avatar
Vitor Sessak committed
114
    sum = av_clipf(sum, 0, 60);
Vitor Sessak's avatar
Vitor Sessak committed
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));
Vitor Sessak's avatar
Vitor Sessak committed
119

120
    for (i=0; i < 5; i++)
121
        buffer[i] = codetable[cb_coef][i] * sumsum;
Vitor Sessak's avatar
Vitor Sessak committed
122

123
    sum = avpriv_scalarproduct_float_c(buffer, buffer, 5);
124

125
    sum = FFMAX(sum, 5.0 / (1<<24));
Vitor Sessak's avatar
Vitor Sessak committed
126 127

    /* shift and store */
128
    memmove(gain_block, gain_block + 1, 9 * sizeof(*gain_block));
Vitor Sessak's avatar
Vitor Sessak committed
129

130
    gain_block[9] = 10 * log10(sum) + (10*log10(((1<<24)/5.)) - 32);
Vitor Sessak's avatar
Vitor Sessak committed
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
 *
138 139 140 141
 * @param order   filter order
 * @param n       input length
 * @param non_rec number of non-recursive samples
 * @param out     filter output
142
 * @param hist    pointer to the input history of the filter
Vitor Sessak's avatar
Vitor Sessak committed
143
 * @param out     pointer to the non-recursive part of the output
144 145 146
 * @param out2    pointer to the recursive part of the output
 * @param window  pointer to the windowing function table
 */
147 148
static void do_hybrid_window(RA288Context *ractx,
                             int order, int n, int non_rec, float *out,
Vitor Sessak's avatar
Vitor Sessak committed
149
                             float *hist, float *out2, const float *window)
150
{
151
    int i;
152 153
    float buffer1[MAX_BACKWARD_FILTER_ORDER + 1];
    float buffer2[MAX_BACKWARD_FILTER_ORDER + 1];
154 155 156
    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
157

158 159
    av_assert2(order>=0);

160
    ractx->fdsp->vector_fmul(work, window, hist, FFALIGN(order + n + non_rec, 16));
161

162 163
    convolve(buffer1, work + order    , n      , order);
    convolve(buffer2, work + order + n, non_rec, order);
164

165 166 167
    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
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
 */
177 178
static void backward_filter(RA288Context *ractx,
                            float *hist, float *rec, const float *window,
179 180
                            float *lpc, const float *tab,
                            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)
Nick Kurshev's avatar
Nick Kurshev committed
194
{
195
    AVFrame *frame     = data;
196 197
    const uint8_t *buf = avpkt->data;
    int buf_size = avpkt->size;
198 199
    float *out;
    int i, ret;
200
    RA288Context *ractx = avctx->priv_data;
201
    GetBitContext gb;
Vitor Sessak's avatar
Vitor Sessak committed
202

203 204 205 206
    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);
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);
Vitor Sessak's avatar
Vitor Sessak committed
217

218
    for (i=0; i < RA288_BLOCKS_PER_FRAME; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
219
        float gain = amptable[get_bits(&gb, 3)];
220
        int cb_coef = get_bits(&gb, 6 + (i&1));
Vitor Sessak's avatar
Vitor Sessak committed
221

222
        decode(ractx, gain, cb_coef);
Vitor Sessak's avatar
Vitor Sessak committed
223

224 225
        memcpy(out, &ractx->sp_hist[70 + 36], RA288_BLOCK_SIZE * sizeof(*out));
        out += RA288_BLOCK_SIZE;
Vitor Sessak's avatar
Vitor Sessak committed
226

227
        if ((i & 7) == 3) {
228
            backward_filter(ractx, ractx->sp_hist, ractx->sp_rec, syn_window,
229 230
                            ractx->sp_lpc, syn_bw_tab, 36, 40, 35, 70);

231
            backward_filter(ractx, ractx->gain_hist, ractx->gain_rec, gain_window,
232 233
                            ractx->gain_lpc, gain_bw_tab, 10, 8, 20, 28);
        }
Vitor Sessak's avatar
Vitor Sessak committed
234 235
    }

236
    *got_frame_ptr = 1;
237

238
    return avctx->block_align;
239 240
}

241 242
AVCodec ff_ra_288_decoder = {
    .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,
246 247 248
    .priv_data_size = sizeof(RA288Context),
    .init           = ra288_decode_init,
    .decode         = ra288_decode_frame,
249
    .close          = ra288_decode_close,
250
    .capabilities   = CODEC_CAP_DR1,
251
};