1 /*
2 * TwinVQ decoder
3 * Copyright (c) 2009 Vitor Sessak
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 <math.h>
23 #include <stdint.h>
24
25 #include "libavutil/channel_layout.h"
26 #include "avcodec.h"
27 #include "get_bits.h"
28 #include "internal.h"
29 #include "twinvq.h"
30 #include "metasound_data.h"
31 #include "twinvq_data.h"
32
33 static const TwinVQModeTab mode_08_08 = {
34 {
35 { 8, bark_tab_s08_64, 10, tab.fcb08s, 1, 5, tab.cb0808s0, tab.cb0808s1, 18 },
36 { 2, bark_tab_m08_256, 20, tab.fcb08m, 2, 5, tab.cb0808m0, tab.cb0808m1, 16 },
37 { 1, bark_tab_l08_512, 30, tab.fcb08l, 3, 6, tab.cb0808l0, tab.cb0808l1, 17 }
38 },
39 512, 12, ff_metasound_lsp8, 1, 5, 3, 3, tab.shape08, 8, 28, 20, 6, 40
40 };
41
42 static const TwinVQModeTab mode_11_08 = {
43 {
44 { 8, bark_tab_s11_64, 10, tab.fcb11s, 1, 5, tab.cb1108s0, tab.cb1108s1, 29 },
45 { 2, bark_tab_m11_256, 20, tab.fcb11m, 2, 5, tab.cb1108m0, tab.cb1108m1, 24 },
46 { 1, bark_tab_l11_512, 30, tab.fcb11l, 3, 6, tab.cb1108l0, tab.cb1108l1, 27 }
47 },
48 512, 16, ff_metasound_lsp11, 1, 6, 4, 3, tab.shape11, 9, 36, 30, 7, 90
49 };
50
51 static const TwinVQModeTab mode_11_10 = {
52 {
53 { 8, bark_tab_s11_64, 10, tab.fcb11s, 1, 5, tab.cb1110s0, tab.cb1110s1, 21 },
54 { 2, bark_tab_m11_256, 20, tab.fcb11m, 2, 5, tab.cb1110m0, tab.cb1110m1, 18 },
55 { 1, bark_tab_l11_512, 30, tab.fcb11l, 3, 6, tab.cb1110l0, tab.cb1110l1, 20 }
56 },
57 512, 16, ff_metasound_lsp11, 1, 6, 4, 3, tab.shape11, 9, 36, 30, 7, 90
58 };
59
60 static const TwinVQModeTab mode_16_16 = {
61 {
62 { 8, bark_tab_s16_128, 10, tab.fcb16s, 1, 5, tab.cb1616s0, tab.cb1616s1, 16 },
63 { 2, bark_tab_m16_512, 20, tab.fcb16m, 2, 5, tab.cb1616m0, tab.cb1616m1, 15 },
64 { 1, bark_tab_l16_1024, 30, tab.fcb16l, 3, 6, tab.cb1616l0, tab.cb1616l1, 16 }
65 },
66 1024, 16, ff_metasound_lsp16, 1, 6, 4, 3, tab.shape16, 9, 56, 60, 7, 180
67 };
68
69 static const TwinVQModeTab mode_22_20 = {
70 {
71 { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2220s0, tab.cb2220s1, 18 },
72 { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2220m0, tab.cb2220m1, 17 },
73 { 1, bark_tab_l22_1024, 32, tab.fcb22l_1, 4, 6, tab.cb2220l0, tab.cb2220l1, 18 }
74 },
75 1024, 16, ff_metasound_lsp22, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
76 };
77
78 static const TwinVQModeTab mode_22_24 = {
79 {
80 { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2224s0, tab.cb2224s1, 15 },
81 { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2224m0, tab.cb2224m1, 14 },
82 { 1, bark_tab_l22_1024, 32, tab.fcb22l_1, 4, 6, tab.cb2224l0, tab.cb2224l1, 15 }
83 },
84 1024, 16, ff_metasound_lsp22, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
85 };
86
87 static const TwinVQModeTab mode_22_32 = {
88 {
89 { 4, bark_tab_s22_128, 10, tab.fcb22s_2, 1, 6, tab.cb2232s0, tab.cb2232s1, 11 },
90 { 2, bark_tab_m22_256, 20, tab.fcb22m_2, 2, 6, tab.cb2232m0, tab.cb2232m1, 11 },
91 { 1, bark_tab_l22_512, 32, tab.fcb22l_2, 4, 6, tab.cb2232l0, tab.cb2232l1, 12 }
92 },
93 512, 16, tab.lsp22_2, 1, 6, 4, 4, tab.shape22_2, 9, 56, 36, 7, 72
94 };
95
96 static const TwinVQModeTab mode_44_40 = {
97 {
98 { 16, bark_tab_s44_128, 10, tab.fcb44s, 1, 6, tab.cb4440s0, tab.cb4440s1, 18 },
99 { 4, bark_tab_m44_512, 20, tab.fcb44m, 2, 6, tab.cb4440m0, tab.cb4440m1, 17 },
100 { 1, bark_tab_l44_2048, 40, tab.fcb44l, 4, 6, tab.cb4440l0, tab.cb4440l1, 17 }
101 },
102 2048, 20, ff_metasound_lsp44, 1, 6, 4, 4, tab.shape44, 9, 84, 54, 7, 432
103 };
104
105 static const TwinVQModeTab mode_44_48 = {
106 {
107 { 16, bark_tab_s44_128, 10, tab.fcb44s, 1, 6, tab.cb4448s0, tab.cb4448s1, 15 },
108 { 4, bark_tab_m44_512, 20, tab.fcb44m, 2, 6, tab.cb4448m0, tab.cb4448m1, 14 },
109 { 1, bark_tab_l44_2048, 40, tab.fcb44l, 4, 6, tab.cb4448l0, tab.cb4448l1, 14 }
110 },
111 2048, 20, ff_metasound_lsp44, 1, 6, 4, 4, tab.shape44, 9, 84, 54, 7, 432
112 };
113
114 /**
115 * Evaluate a * b / 400 rounded to the nearest integer. When, for example,
116 * a * b == 200 and the nearest integer is ill-defined, use a table to emulate
117 * the following broken float-based implementation used by the binary decoder:
118 *
119 * @code
120 * static int very_broken_op(int a, int b)
121 * {
122 * static float test; // Ugh, force gcc to do the division first...
123 *
124 * test = a / 400.0;
125 * return b * test + 0.5;
126 * }
127 * @endcode
128 *
129 * @note if this function is replaced by just ROUNDED_DIV(a * b, 400.0), the
130 * stddev between the original file (before encoding with Yamaha encoder) and
131 * the decoded output increases, which leads one to believe that the encoder
132 * expects exactly this broken calculation.
133 */
very_broken_op(int a,int b)134 static int very_broken_op(int a, int b)
135 {
136 int x = a * b + 200;
137 int size;
138 const uint8_t *rtab;
139
140 if (x % 400 || b % 5)
141 return x / 400;
142
143 x /= 400;
144
145 size = tabs[b / 5].size;
146 rtab = tabs[b / 5].tab;
147 return x - rtab[size * av_log2(2 * (x - 1) / size) + (x - 1) % size];
148 }
149
150 /**
151 * Sum to data a periodic peak of a given period, width and shape.
152 *
153 * @param period the period of the peak divided by 400.0
154 */
add_peak(int period,int width,const float * shape,float ppc_gain,float * speech,int len)155 static void add_peak(int period, int width, const float *shape,
156 float ppc_gain, float *speech, int len)
157 {
158 int i, j;
159
160 const float *shape_end = shape + len;
161 int center;
162
163 // First peak centered around zero
164 for (i = 0; i < width / 2; i++)
165 speech[i] += ppc_gain * *shape++;
166
167 for (i = 1; i < ROUNDED_DIV(len, width); i++) {
168 center = very_broken_op(period, i);
169 for (j = -width / 2; j < (width + 1) / 2; j++)
170 speech[j + center] += ppc_gain * *shape++;
171 }
172
173 // For the last block, be careful not to go beyond the end of the buffer
174 center = very_broken_op(period, i);
175 for (j = -width / 2; j < (width + 1) / 2 && shape < shape_end; j++)
176 speech[j + center] += ppc_gain * *shape++;
177 }
178
decode_ppc(TwinVQContext * tctx,int period_coef,int g_coef,const float * shape,float * speech)179 static void decode_ppc(TwinVQContext *tctx, int period_coef, int g_coef,
180 const float *shape, float *speech)
181 {
182 const TwinVQModeTab *mtab = tctx->mtab;
183 int isampf = tctx->avctx->sample_rate / 1000;
184 int ibps = tctx->avctx->bit_rate / (1000 * tctx->avctx->channels);
185 int min_period = ROUNDED_DIV(40 * 2 * mtab->size, isampf);
186 int max_period = ROUNDED_DIV(40 * 2 * mtab->size * 6, isampf);
187 int period_range = max_period - min_period;
188 float pgain_step = 25000.0 / ((1 << mtab->pgain_bit) - 1);
189 float ppc_gain = 1.0 / 8192 *
190 twinvq_mulawinv(pgain_step * g_coef +
191 pgain_step / 2,
192 25000.0, TWINVQ_PGAIN_MU);
193
194 // This is actually the period multiplied by 400. It is just linearly coded
195 // between its maximum and minimum value.
196 int period = min_period +
197 ROUNDED_DIV(period_coef * period_range,
198 (1 << mtab->ppc_period_bit) - 1);
199 int width;
200
201 if (isampf == 22 && ibps == 32) {
202 // For some unknown reason, NTT decided to code this case differently...
203 width = ROUNDED_DIV((period + 800) * mtab->peak_per2wid,
204 400 * mtab->size);
205 } else
206 width = period * mtab->peak_per2wid / (400 * mtab->size);
207
208 add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len);
209 }
210
dec_bark_env(TwinVQContext * tctx,const uint8_t * in,int use_hist,int ch,float * out,float gain,enum TwinVQFrameType ftype)211 static void dec_bark_env(TwinVQContext *tctx, const uint8_t *in, int use_hist,
212 int ch, float *out, float gain,
213 enum TwinVQFrameType ftype)
214 {
215 const TwinVQModeTab *mtab = tctx->mtab;
216 int i, j;
217 float *hist = tctx->bark_hist[ftype][ch];
218 float val = ((const float []) { 0.4, 0.35, 0.28 })[ftype];
219 int bark_n_coef = mtab->fmode[ftype].bark_n_coef;
220 int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef;
221 int idx = 0;
222
223 for (i = 0; i < fw_cb_len; i++)
224 for (j = 0; j < bark_n_coef; j++, idx++) {
225 float tmp2 = mtab->fmode[ftype].bark_cb[fw_cb_len * in[j] + i] *
226 (1.0 / 4096);
227 float st = use_hist ? (1.0 - val) * tmp2 + val * hist[idx] + 1.0
228 : tmp2 + 1.0;
229
230 hist[idx] = tmp2;
231 if (st < -1.0)
232 st = 1.0;
233
234 twinvq_memset_float(out, st * gain, mtab->fmode[ftype].bark_tab[idx]);
235 out += mtab->fmode[ftype].bark_tab[idx];
236 }
237 }
238
read_cb_data(TwinVQContext * tctx,GetBitContext * gb,uint8_t * dst,enum TwinVQFrameType ftype)239 static void read_cb_data(TwinVQContext *tctx, GetBitContext *gb,
240 uint8_t *dst, enum TwinVQFrameType ftype)
241 {
242 int i;
243
244 for (i = 0; i < tctx->n_div[ftype]; i++) {
245 int bs_second_part = (i >= tctx->bits_main_spec_change[ftype]);
246
247 *dst++ = get_bits(gb, tctx->bits_main_spec[0][ftype][bs_second_part]);
248 *dst++ = get_bits(gb, tctx->bits_main_spec[1][ftype][bs_second_part]);
249 }
250 }
251
twinvq_read_bitstream(AVCodecContext * avctx,TwinVQContext * tctx,const uint8_t * buf,int buf_size)252 static int twinvq_read_bitstream(AVCodecContext *avctx, TwinVQContext *tctx,
253 const uint8_t *buf, int buf_size)
254 {
255 TwinVQFrameData *bits = &tctx->bits[0];
256 const TwinVQModeTab *mtab = tctx->mtab;
257 int channels = tctx->avctx->channels;
258 int sub;
259 GetBitContext gb;
260 int i, j, k, ret;
261
262 if ((ret = init_get_bits8(&gb, buf, buf_size)) < 0)
263 return ret;
264 skip_bits(&gb, get_bits(&gb, 8));
265
266 bits->window_type = get_bits(&gb, TWINVQ_WINDOW_TYPE_BITS);
267
268 if (bits->window_type > 8) {
269 av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
270 return AVERROR_INVALIDDATA;
271 }
272
273 bits->ftype = ff_twinvq_wtype_to_ftype_table[tctx->bits[0].window_type];
274
275 sub = mtab->fmode[bits->ftype].sub;
276
277 read_cb_data(tctx, &gb, bits->main_coeffs, bits->ftype);
278
279 for (i = 0; i < channels; i++)
280 for (j = 0; j < sub; j++)
281 for (k = 0; k < mtab->fmode[bits->ftype].bark_n_coef; k++)
282 bits->bark1[i][j][k] =
283 get_bits(&gb, mtab->fmode[bits->ftype].bark_n_bit);
284
285 for (i = 0; i < channels; i++)
286 for (j = 0; j < sub; j++)
287 bits->bark_use_hist[i][j] = get_bits1(&gb);
288
289 if (bits->ftype == TWINVQ_FT_LONG) {
290 for (i = 0; i < channels; i++)
291 bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS);
292 } else {
293 for (i = 0; i < channels; i++) {
294 bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS);
295 for (j = 0; j < sub; j++)
296 bits->sub_gain_bits[i * sub + j] = get_bits(&gb,
297 TWINVQ_SUB_GAIN_BITS);
298 }
299 }
300
301 for (i = 0; i < channels; i++) {
302 bits->lpc_hist_idx[i] = get_bits(&gb, mtab->lsp_bit0);
303 bits->lpc_idx1[i] = get_bits(&gb, mtab->lsp_bit1);
304
305 for (j = 0; j < mtab->lsp_split; j++)
306 bits->lpc_idx2[i][j] = get_bits(&gb, mtab->lsp_bit2);
307 }
308
309 if (bits->ftype == TWINVQ_FT_LONG) {
310 read_cb_data(tctx, &gb, bits->ppc_coeffs, 3);
311 for (i = 0; i < channels; i++) {
312 bits->p_coef[i] = get_bits(&gb, mtab->ppc_period_bit);
313 bits->g_coef[i] = get_bits(&gb, mtab->pgain_bit);
314 }
315 }
316
317 return (get_bits_count(&gb) + 7) / 8;
318 }
319
twinvq_decode_init(AVCodecContext * avctx)320 static av_cold int twinvq_decode_init(AVCodecContext *avctx)
321 {
322 int isampf, ibps;
323 TwinVQContext *tctx = avctx->priv_data;
324
325 if (!avctx->extradata || avctx->extradata_size < 12) {
326 av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n");
327 return AVERROR_INVALIDDATA;
328 }
329 avctx->channels = AV_RB32(avctx->extradata) + 1;
330 avctx->bit_rate = AV_RB32(avctx->extradata + 4) * 1000;
331 isampf = AV_RB32(avctx->extradata + 8);
332
333 if (isampf < 8 || isampf > 44) {
334 av_log(avctx, AV_LOG_ERROR, "Unsupported sample rate\n");
335 return AVERROR_INVALIDDATA;
336 }
337 switch (isampf) {
338 case 44:
339 avctx->sample_rate = 44100;
340 break;
341 case 22:
342 avctx->sample_rate = 22050;
343 break;
344 case 11:
345 avctx->sample_rate = 11025;
346 break;
347 default:
348 avctx->sample_rate = isampf * 1000;
349 break;
350 }
351
352 if (avctx->channels <= 0 || avctx->channels > TWINVQ_CHANNELS_MAX) {
353 av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n",
354 avctx->channels);
355 return -1;
356 }
357 avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO
358 : AV_CH_LAYOUT_STEREO;
359
360 ibps = avctx->bit_rate / (1000 * avctx->channels);
361 if (ibps < 8 || ibps > 48) {
362 av_log(avctx, AV_LOG_ERROR, "Bad bitrate per channel value %d\n", ibps);
363 return AVERROR_INVALIDDATA;
364 }
365
366 switch ((isampf << 8) + ibps) {
367 case (8 << 8) + 8:
368 tctx->mtab = &mode_08_08;
369 break;
370 case (11 << 8) + 8:
371 tctx->mtab = &mode_11_08;
372 break;
373 case (11 << 8) + 10:
374 tctx->mtab = &mode_11_10;
375 break;
376 case (16 << 8) + 16:
377 tctx->mtab = &mode_16_16;
378 break;
379 case (22 << 8) + 20:
380 tctx->mtab = &mode_22_20;
381 break;
382 case (22 << 8) + 24:
383 tctx->mtab = &mode_22_24;
384 break;
385 case (22 << 8) + 32:
386 tctx->mtab = &mode_22_32;
387 break;
388 case (44 << 8) + 40:
389 tctx->mtab = &mode_44_40;
390 break;
391 case (44 << 8) + 48:
392 tctx->mtab = &mode_44_48;
393 break;
394 default:
395 av_log(avctx, AV_LOG_ERROR,
396 "This version does not support %d kHz - %d kbit/s/ch mode.\n",
397 isampf, isampf);
398 return -1;
399 }
400
401 tctx->codec = TWINVQ_CODEC_VQF;
402 tctx->read_bitstream = twinvq_read_bitstream;
403 tctx->dec_bark_env = dec_bark_env;
404 tctx->decode_ppc = decode_ppc;
405 tctx->frame_size = avctx->bit_rate * tctx->mtab->size
406 / avctx->sample_rate + 8;
407 tctx->is_6kbps = 0;
408 if (avctx->block_align && avctx->block_align * 8LL / tctx->frame_size > 1) {
409 av_log(avctx, AV_LOG_ERROR,
410 "VQF TwinVQ should have only one frame per packet\n");
411 return AVERROR_INVALIDDATA;
412 }
413
414 return ff_twinvq_decode_init(avctx);
415 }
416
417 AVCodec ff_twinvq_decoder = {
418 .name = "twinvq",
419 .long_name = NULL_IF_CONFIG_SMALL("VQF TwinVQ"),
420 .type = AVMEDIA_TYPE_AUDIO,
421 .id = AV_CODEC_ID_TWINVQ,
422 .priv_data_size = sizeof(TwinVQContext),
423 .init = twinvq_decode_init,
424 .close = ff_twinvq_decode_close,
425 .decode = ff_twinvq_decode_frame,
426 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_CHANNEL_CONF,
427 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
428 AV_SAMPLE_FMT_NONE },
429 };
430