ra288.c 7.57 KB
Newer Older
1 2 3 4
/*
 * RealAudio 2.0 (28.8K)
 * Copyright (c) 2003 the ffmpeg project
 *
5
 * This file is part of Libav.
6
 *
7
 * Libav 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
 * Libav 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 Libav; 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 {
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
static av_cold int ra288_decode_init(AVCodecContext *avctx)
{
64
    RA288Context *ractx = avctx->priv_data;
65 66 67 68 69

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

70
    avpriv_float_dsp_init(&ractx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
71

72 73 74
    return 0;
}

75 76 77
static void convolve(float *tgt, const float *src, int len, int n)
{
    for (; n >= 0; n--)
78
        tgt[n] = avpriv_scalarproduct_float_c(src, src - n, len);
79 80 81

}

82
static void decode(RA288Context *ractx, float gain, int cb_coef)
83
{
84
    int i;
85 86
    double sumsum;
    float sum, buffer[5];
87 88
    float *block = ractx->sp_hist + 70 + 36; // current block
    float *gain_block = ractx->gain_hist + 28;
89

90
    memmove(ractx->sp_hist + 70, ractx->sp_hist + 75, 36*sizeof(*block));
91

92
    /* block 46 of G.728 spec */
93
    sum = 32.0;
94
    for (i=0; i < 10; i++)
95
        sum -= gain_block[9-i] * ractx->gain_lpc[i];
96

97
    /* block 47 of G.728 spec */
Vitor Sessak's avatar
Vitor Sessak committed
98
    sum = av_clipf(sum, 0, 60);
Vitor Sessak's avatar
Vitor Sessak committed
99

100
    /* block 48 of G.728 spec */
101
    /* exp(sum * 0.1151292546497) == pow(10.0,sum/20) */
102
    sumsum = exp(sum * 0.1151292546497) * gain * (1.0/(1<<23));
Vitor Sessak's avatar
Vitor Sessak committed
103

104
    for (i=0; i < 5; i++)
105
        buffer[i] = codetable[cb_coef][i] * sumsum;
Vitor Sessak's avatar
Vitor Sessak committed
106

107
    sum = avpriv_scalarproduct_float_c(buffer, buffer, 5) * ((1 << 24) / 5.0);
108 109

    sum = FFMAX(sum, 1);
Vitor Sessak's avatar
Vitor Sessak committed
110 111

    /* shift and store */
112
    memmove(gain_block, gain_block + 1, 9 * sizeof(*gain_block));
Vitor Sessak's avatar
Vitor Sessak committed
113

114
    gain_block[9] = 10 * log10(sum) - 32;
Vitor Sessak's avatar
Vitor Sessak committed
115

116
    ff_celp_lp_synthesis_filterf(block, ractx->sp_lpc, buffer, 5, 36);
117 118
}

119
/**
120
 * Hybrid window filtering, see blocks 36 and 49 of the G.728 specification.
121
 *
122 123 124 125
 * @param order   filter order
 * @param n       input length
 * @param non_rec number of non-recursive samples
 * @param out     filter output
126
 * @param hist    pointer to the input history of the filter
Vitor Sessak's avatar
Vitor Sessak committed
127
 * @param out     pointer to the non-recursive part of the output
128 129 130
 * @param out2    pointer to the recursive part of the output
 * @param window  pointer to the windowing function table
 */
131 132
static void do_hybrid_window(RA288Context *ractx,
                             int order, int n, int non_rec, float *out,
Vitor Sessak's avatar
Vitor Sessak committed
133
                             float *hist, float *out2, const float *window)
134
{
135
    int i;
136 137
    float buffer1[MAX_BACKWARD_FILTER_ORDER + 1];
    float buffer2[MAX_BACKWARD_FILTER_ORDER + 1];
138 139 140
    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
141

142
    ractx->fdsp.vector_fmul(work, window, hist, FFALIGN(order + n + non_rec, 16));
143

144 145
    convolve(buffer1, work + order    , n      , order);
    convolve(buffer2, work + order + n, non_rec, order);
146

147 148 149
    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
150
    }
151

152
    /* Multiply by the white noise correcting factor (WNCF). */
153
    *out *= 257.0 / 256.0;
154 155
}

156
/**
157
 * Backward synthesis filter, find the LPC coefficients from past speech data.
158
 */
159 160
static void backward_filter(RA288Context *ractx,
                            float *hist, float *rec, const float *window,
161 162
                            float *lpc, const float *tab,
                            int order, int n, int non_rec, int move_size)
163
{
164
    float temp[MAX_BACKWARD_FILTER_ORDER+1];
165

166
    do_hybrid_window(ractx, order, n, non_rec, temp, hist, rec, window);
167

168
    if (!compute_lpc_coefs(temp, order, lpc, 0, 1, 1))
169
        ractx->fdsp.vector_fmul(lpc, lpc, tab, FFALIGN(order, 16));
170

171
    memmove(hist, hist + n, move_size*sizeof(*hist));
172 173
}

174
static int ra288_decode_frame(AVCodecContext * avctx, void *data,
175
                              int *got_frame_ptr, AVPacket *avpkt)
Nick Kurshev's avatar
Nick Kurshev committed
176
{
177
    AVFrame *frame     = data;
178 179
    const uint8_t *buf = avpkt->data;
    int buf_size = avpkt->size;
180 181
    float *out;
    int i, ret;
182
    RA288Context *ractx = avctx->priv_data;
183
    GetBitContext gb;
Vitor Sessak's avatar
Vitor Sessak committed
184

185 186 187 188
    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);
189
        return AVERROR_INVALIDDATA;
190 191
    }

192
    /* get output buffer */
193
    frame->nb_samples = RA288_BLOCK_SIZE * RA288_BLOCKS_PER_FRAME;
194
    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
195 196
        av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
        return ret;
197
    }
198
    out = (float *)frame->data[0];
199

200
    init_get_bits(&gb, buf, avctx->block_align * 8);
Vitor Sessak's avatar
Vitor Sessak committed
201

202
    for (i=0; i < RA288_BLOCKS_PER_FRAME; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
203
        float gain = amptable[get_bits(&gb, 3)];
204
        int cb_coef = get_bits(&gb, 6 + (i&1));
Vitor Sessak's avatar
Vitor Sessak committed
205

206
        decode(ractx, gain, cb_coef);
Vitor Sessak's avatar
Vitor Sessak committed
207

208 209
        memcpy(out, &ractx->sp_hist[70 + 36], RA288_BLOCK_SIZE * sizeof(*out));
        out += RA288_BLOCK_SIZE;
Vitor Sessak's avatar
Vitor Sessak committed
210

211
        if ((i & 7) == 3) {
212
            backward_filter(ractx, ractx->sp_hist, ractx->sp_rec, syn_window,
213 214
                            ractx->sp_lpc, syn_bw_tab, 36, 40, 35, 70);

215
            backward_filter(ractx, ractx->gain_hist, ractx->gain_rec, gain_window,
216 217
                            ractx->gain_lpc, gain_bw_tab, 10, 8, 20, 28);
        }
Vitor Sessak's avatar
Vitor Sessak committed
218 219
    }

220
    *got_frame_ptr = 1;
221

222
    return avctx->block_align;
223 224
}

225 226
AVCodec ff_ra_288_decoder = {
    .name           = "real_288",
227
    .long_name      = NULL_IF_CONFIG_SMALL("RealAudio 2.0 (28.8K)"),
228
    .type           = AVMEDIA_TYPE_AUDIO,
229
    .id             = AV_CODEC_ID_RA_288,
230 231 232
    .priv_data_size = sizeof(RA288Context),
    .init           = ra288_decode_init,
    .decode         = ra288_decode_frame,
233
    .capabilities   = CODEC_CAP_DR1,
234
};