xref: /third_party/ffmpeg/libavcodec/ra288.c (revision cabdff1a)
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 "libavutil/channel_layout.h"
23#include "libavutil/float_dsp.h"
24#include "libavutil/internal.h"
25#include "libavutil/mem_internal.h"
26
27#define BITSTREAM_READER_LE
28#include "avcodec.h"
29#include "celp_filters.h"
30#include "codec_internal.h"
31#include "get_bits.h"
32#include "internal.h"
33#include "lpc.h"
34#include "ra288.h"
35
36#define MAX_BACKWARD_FILTER_ORDER  36
37#define MAX_BACKWARD_FILTER_LEN    40
38#define MAX_BACKWARD_FILTER_NONREC 35
39
40#define RA288_BLOCK_SIZE        5
41#define RA288_BLOCKS_PER_FRAME 32
42
43typedef struct RA288Context {
44    void (*vector_fmul)(float *dst, const float *src0, const float *src1,
45                        int len);
46    DECLARE_ALIGNED(32, float,   sp_lpc)[FFALIGN(36, 16)];   ///< LPC coefficients for speech data (spec: A)
47    DECLARE_ALIGNED(32, float, gain_lpc)[FFALIGN(10, 16)];   ///< LPC coefficients for gain        (spec: GB)
48
49    /** speech data history                                      (spec: SB).
50     *  Its first 70 coefficients are updated only at backward filtering.
51     */
52    float sp_hist[111];
53
54    /// speech part of the gain autocorrelation                  (spec: REXP)
55    float sp_rec[37];
56
57    /** log-gain history                                         (spec: SBLG).
58     *  Its first 28 coefficients are updated only at backward filtering.
59     */
60    float gain_hist[38];
61
62    /// recursive part of the gain autocorrelation               (spec: REXPLG)
63    float gain_rec[11];
64} RA288Context;
65
66static av_cold int ra288_decode_init(AVCodecContext *avctx)
67{
68    RA288Context *ractx = avctx->priv_data;
69    AVFloatDSPContext *fdsp;
70
71    av_channel_layout_uninit(&avctx->ch_layout);
72    avctx->ch_layout      = (AVChannelLayout)AV_CHANNEL_LAYOUT_MONO;
73    avctx->sample_fmt     = AV_SAMPLE_FMT_FLT;
74
75    if (avctx->block_align != 38) {
76        av_log(avctx, AV_LOG_ERROR, "unsupported block align\n");
77        return AVERROR_PATCHWELCOME;
78    }
79
80    fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
81    if (!fdsp)
82        return AVERROR(ENOMEM);
83    ractx->vector_fmul = fdsp->vector_fmul;
84    av_free(fdsp);
85
86    return 0;
87}
88
89static void convolve(float *tgt, const float *src, int len, int n)
90{
91    for (; n >= 0; n--)
92        tgt[n] = avpriv_scalarproduct_float_c(src, src - n, len);
93
94}
95
96static void decode(RA288Context *ractx, float gain, int cb_coef)
97{
98    int i;
99    double sumsum;
100    float sum, buffer[5];
101    float *block = ractx->sp_hist + 70 + 36; // current block
102    float *gain_block = ractx->gain_hist + 28;
103
104    memmove(ractx->sp_hist + 70, ractx->sp_hist + 75, 36*sizeof(*block));
105
106    /* block 46 of G.728 spec */
107    sum = 32.0;
108    for (i=0; i < 10; i++)
109        sum -= gain_block[9-i] * ractx->gain_lpc[i];
110
111    /* block 47 of G.728 spec */
112    sum = av_clipf(sum, 0, 60);
113
114    /* block 48 of G.728 spec */
115    /* exp(sum * 0.1151292546497) == pow(10.0,sum/20) */
116    sumsum = exp(sum * 0.1151292546497) * gain * (1.0/(1<<23));
117
118    for (i=0; i < 5; i++)
119        buffer[i] = codetable[cb_coef][i] * sumsum;
120
121    sum = avpriv_scalarproduct_float_c(buffer, buffer, 5);
122
123    sum = FFMAX(sum, 5.0 / (1<<24));
124
125    /* shift and store */
126    memmove(gain_block, gain_block + 1, 9 * sizeof(*gain_block));
127
128    gain_block[9] = 10 * log10(sum) + (10*log10(((1<<24)/5.)) - 32);
129
130    ff_celp_lp_synthesis_filterf(block, ractx->sp_lpc, buffer, 5, 36);
131}
132
133/**
134 * Hybrid window filtering, see blocks 36 and 49 of the G.728 specification.
135 *
136 * @param order   filter order
137 * @param n       input length
138 * @param non_rec number of non-recursive samples
139 * @param out     filter output
140 * @param hist    pointer to the input history of the filter
141 * @param out     pointer to the non-recursive part of the output
142 * @param out2    pointer to the recursive part of the output
143 * @param window  pointer to the windowing function table
144 */
145static void do_hybrid_window(RA288Context *ractx,
146                             int order, int n, int non_rec, float *out,
147                             float *hist, float *out2, const float *window)
148{
149    int i;
150    float buffer1[MAX_BACKWARD_FILTER_ORDER + 1];
151    float buffer2[MAX_BACKWARD_FILTER_ORDER + 1];
152    LOCAL_ALIGNED(32, float, work, [FFALIGN(MAX_BACKWARD_FILTER_ORDER +
153                                            MAX_BACKWARD_FILTER_LEN   +
154                                            MAX_BACKWARD_FILTER_NONREC, 16)]);
155
156    av_assert2(order>=0);
157
158    ractx->vector_fmul(work, window, hist, FFALIGN(order + n + non_rec, 16));
159
160    convolve(buffer1, work + order    , n      , order);
161    convolve(buffer2, work + order + n, non_rec, order);
162
163    for (i=0; i <= order; i++) {
164        out2[i] = out2[i] * 0.5625 + buffer1[i];
165        out [i] = out2[i]          + buffer2[i];
166    }
167
168    /* Multiply by the white noise correcting factor (WNCF). */
169    *out *= 257.0 / 256.0;
170}
171
172/**
173 * Backward synthesis filter, find the LPC coefficients from past speech data.
174 */
175static void backward_filter(RA288Context *ractx,
176                            float *hist, float *rec, const float *window,
177                            float *lpc, const float *tab,
178                            int order, int n, int non_rec, int move_size)
179{
180    float temp[MAX_BACKWARD_FILTER_ORDER+1];
181
182    do_hybrid_window(ractx, order, n, non_rec, temp, hist, rec, window);
183
184    if (!compute_lpc_coefs(temp, order, lpc, 0, 1, 1))
185        ractx->vector_fmul(lpc, lpc, tab, FFALIGN(order, 16));
186
187    memmove(hist, hist + n, move_size*sizeof(*hist));
188}
189
190static int ra288_decode_frame(AVCodecContext * avctx, AVFrame *frame,
191                              int *got_frame_ptr, AVPacket *avpkt)
192{
193    const uint8_t *buf = avpkt->data;
194    int buf_size = avpkt->size;
195    float *out;
196    int i, ret;
197    RA288Context *ractx = avctx->priv_data;
198    GetBitContext gb;
199
200    if (buf_size < avctx->block_align) {
201        av_log(avctx, AV_LOG_ERROR,
202               "Error! Input buffer is too small [%d<%d]\n",
203               buf_size, avctx->block_align);
204        return AVERROR_INVALIDDATA;
205    }
206
207    ret = init_get_bits8(&gb, buf, avctx->block_align);
208    if (ret < 0)
209        return ret;
210
211    /* get output buffer */
212    frame->nb_samples = RA288_BLOCK_SIZE * RA288_BLOCKS_PER_FRAME;
213    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
214        return ret;
215    out = (float *)frame->data[0];
216
217    for (i=0; i < RA288_BLOCKS_PER_FRAME; i++) {
218        float gain = amptable[get_bits(&gb, 3)];
219        int cb_coef = get_bits(&gb, 6 + (i&1));
220
221        decode(ractx, gain, cb_coef);
222
223        memcpy(out, &ractx->sp_hist[70 + 36], RA288_BLOCK_SIZE * sizeof(*out));
224        out += RA288_BLOCK_SIZE;
225
226        if ((i & 7) == 3) {
227            backward_filter(ractx, ractx->sp_hist, ractx->sp_rec, syn_window,
228                            ractx->sp_lpc, syn_bw_tab, 36, 40, 35, 70);
229
230            backward_filter(ractx, ractx->gain_hist, ractx->gain_rec, gain_window,
231                            ractx->gain_lpc, gain_bw_tab, 10, 8, 20, 28);
232        }
233    }
234
235    *got_frame_ptr = 1;
236
237    return avctx->block_align;
238}
239
240const FFCodec ff_ra_288_decoder = {
241    .p.name         = "real_288",
242    .p.long_name    = NULL_IF_CONFIG_SMALL("RealAudio 2.0 (28.8K)"),
243    .p.type         = AVMEDIA_TYPE_AUDIO,
244    .p.id           = AV_CODEC_ID_RA_288,
245    .priv_data_size = sizeof(RA288Context),
246    .init           = ra288_decode_init,
247    FF_CODEC_DECODE_CB(ra288_decode_frame),
248    .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_CHANNEL_CONF,
249    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE,
250};
251