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 "get_bits.h"
31 #include "internal.h"
32 #include "lpc.h"
33 #include "ra288.h"
34
35 #define MAX_BACKWARD_FILTER_ORDER 36
36 #define MAX_BACKWARD_FILTER_LEN 40
37 #define MAX_BACKWARD_FILTER_NONREC 35
38
39 #define RA288_BLOCK_SIZE 5
40 #define RA288_BLOCKS_PER_FRAME 32
41
42 typedef struct RA288Context {
43 void (*vector_fmul)(float *dst, const float *src0, const float *src1,
44 int len);
45 DECLARE_ALIGNED(32, float, sp_lpc)[FFALIGN(36, 16)]; ///< LPC coefficients for speech data (spec: A)
46 DECLARE_ALIGNED(32, float, gain_lpc)[FFALIGN(10, 16)]; ///< LPC coefficients for gain (spec: GB)
47
48 /** speech data history (spec: SB).
49 * Its first 70 coefficients are updated only at backward filtering.
50 */
51 float sp_hist[111];
52
53 /// speech part of the gain autocorrelation (spec: REXP)
54 float sp_rec[37];
55
56 /** log-gain history (spec: SBLG).
57 * Its first 28 coefficients are updated only at backward filtering.
58 */
59 float gain_hist[38];
60
61 /// recursive part of the gain autocorrelation (spec: REXPLG)
62 float gain_rec[11];
63 } RA288Context;
64
ra288_decode_init(AVCodecContext * avctx)65 static av_cold int ra288_decode_init(AVCodecContext *avctx)
66 {
67 RA288Context *ractx = avctx->priv_data;
68 AVFloatDSPContext *fdsp;
69
70 avctx->channels = 1;
71 avctx->channel_layout = AV_CH_LAYOUT_MONO;
72 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
73
74 if (avctx->block_align != 38) {
75 av_log(avctx, AV_LOG_ERROR, "unsupported block align\n");
76 return AVERROR_PATCHWELCOME;
77 }
78
79 fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
80 if (!fdsp)
81 return AVERROR(ENOMEM);
82 ractx->vector_fmul = fdsp->vector_fmul;
83 av_free(fdsp);
84
85 return 0;
86 }
87
convolve(float * tgt,const float * src,int len,int n)88 static void convolve(float *tgt, const float *src, int len, int n)
89 {
90 for (; n >= 0; n--)
91 tgt[n] = avpriv_scalarproduct_float_c(src, src - n, len);
92
93 }
94
decode(RA288Context * ractx,float gain,int cb_coef)95 static void decode(RA288Context *ractx, float gain, int cb_coef)
96 {
97 int i;
98 double sumsum;
99 float sum, buffer[5];
100 float *block = ractx->sp_hist + 70 + 36; // current block
101 float *gain_block = ractx->gain_hist + 28;
102
103 memmove(ractx->sp_hist + 70, ractx->sp_hist + 75, 36*sizeof(*block));
104
105 /* block 46 of G.728 spec */
106 sum = 32.0;
107 for (i=0; i < 10; i++)
108 sum -= gain_block[9-i] * ractx->gain_lpc[i];
109
110 /* block 47 of G.728 spec */
111 sum = av_clipf(sum, 0, 60);
112
113 /* block 48 of G.728 spec */
114 /* exp(sum * 0.1151292546497) == pow(10.0,sum/20) */
115 sumsum = exp(sum * 0.1151292546497) * gain * (1.0/(1<<23));
116
117 for (i=0; i < 5; i++)
118 buffer[i] = codetable[cb_coef][i] * sumsum;
119
120 sum = avpriv_scalarproduct_float_c(buffer, buffer, 5);
121
122 sum = FFMAX(sum, 5.0 / (1<<24));
123
124 /* shift and store */
125 memmove(gain_block, gain_block + 1, 9 * sizeof(*gain_block));
126
127 gain_block[9] = 10 * log10(sum) + (10*log10(((1<<24)/5.)) - 32);
128
129 ff_celp_lp_synthesis_filterf(block, ractx->sp_lpc, buffer, 5, 36);
130 }
131
132 /**
133 * Hybrid window filtering, see blocks 36 and 49 of the G.728 specification.
134 *
135 * @param order filter order
136 * @param n input length
137 * @param non_rec number of non-recursive samples
138 * @param out filter output
139 * @param hist pointer to the input history of the filter
140 * @param out pointer to the non-recursive part of the output
141 * @param out2 pointer to the recursive part of the output
142 * @param window pointer to the windowing function table
143 */
do_hybrid_window(RA288Context * ractx,int order,int n,int non_rec,float * out,float * hist,float * out2,const float * window)144 static void do_hybrid_window(RA288Context *ractx,
145 int order, int n, int non_rec, float *out,
146 float *hist, float *out2, const float *window)
147 {
148 int i;
149 float buffer1[MAX_BACKWARD_FILTER_ORDER + 1];
150 float buffer2[MAX_BACKWARD_FILTER_ORDER + 1];
151 LOCAL_ALIGNED(32, float, work, [FFALIGN(MAX_BACKWARD_FILTER_ORDER +
152 MAX_BACKWARD_FILTER_LEN +
153 MAX_BACKWARD_FILTER_NONREC, 16)]);
154
155 av_assert2(order>=0);
156
157 ractx->vector_fmul(work, window, hist, FFALIGN(order + n + non_rec, 16));
158
159 convolve(buffer1, work + order , n , order);
160 convolve(buffer2, work + order + n, non_rec, order);
161
162 for (i=0; i <= order; i++) {
163 out2[i] = out2[i] * 0.5625 + buffer1[i];
164 out [i] = out2[i] + buffer2[i];
165 }
166
167 /* Multiply by the white noise correcting factor (WNCF). */
168 *out *= 257.0 / 256.0;
169 }
170
171 /**
172 * Backward synthesis filter, find the LPC coefficients from past speech data.
173 */
backward_filter(RA288Context * ractx,float * hist,float * rec,const float * window,float * lpc,const float * tab,int order,int n,int non_rec,int move_size)174 static void backward_filter(RA288Context *ractx,
175 float *hist, float *rec, const float *window,
176 float *lpc, const float *tab,
177 int order, int n, int non_rec, int move_size)
178 {
179 float temp[MAX_BACKWARD_FILTER_ORDER+1];
180
181 do_hybrid_window(ractx, order, n, non_rec, temp, hist, rec, window);
182
183 if (!compute_lpc_coefs(temp, order, lpc, 0, 1, 1))
184 ractx->vector_fmul(lpc, lpc, tab, FFALIGN(order, 16));
185
186 memmove(hist, hist + n, move_size*sizeof(*hist));
187 }
188
ra288_decode_frame(AVCodecContext * avctx,void * data,int * got_frame_ptr,AVPacket * avpkt)189 static int ra288_decode_frame(AVCodecContext * avctx, void *data,
190 int *got_frame_ptr, AVPacket *avpkt)
191 {
192 AVFrame *frame = data;
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
240 AVCodec ff_ra_288_decoder = {
241 .name = "real_288",
242 .long_name = NULL_IF_CONFIG_SMALL("RealAudio 2.0 (28.8K)"),
243 .type = AVMEDIA_TYPE_AUDIO,
244 .id = AV_CODEC_ID_RA_288,
245 .priv_data_size = sizeof(RA288Context),
246 .init = ra288_decode_init,
247 .decode = ra288_decode_frame,
248 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_CHANNEL_CONF,
249 };
250