1/* 2 * AAC Spectral Band Replication decoding functions 3 * Copyright (c) 2008-2009 Robert Swain ( rob opendot cl ) 4 * Copyright (c) 2009-2010 Alex Converse <alex.converse@gmail.com> 5 * 6 * This file is part of Libav. 7 * 8 * Libav is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU Lesser General Public 10 * License as published by the Free Software Foundation; either 11 * version 2.1 of the License, or (at your option) any later version. 12 * 13 * Libav is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 16 * Lesser General Public License for more details. 17 * 18 * You should have received a copy of the GNU Lesser General Public 19 * License along with Libav; if not, write to the Free Software 20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 21 */ 22 23/** 24 * @file 25 * AAC Spectral Band Replication decoding functions 26 * @author Robert Swain ( rob opendot cl ) 27 */ 28 29#include "aac.h" 30#include "sbr.h" 31#include "aacsbr.h" 32#include "aacsbrdata.h" 33#include "fft.h" 34#include "aacps.h" 35#include "libavutil/libm.h" 36 37#include <stdint.h> 38#include <float.h> 39 40#define ENVELOPE_ADJUSTMENT_OFFSET 2 41#define NOISE_FLOOR_OFFSET 6.0f 42 43/** 44 * SBR VLC tables 45 */ 46enum { 47 T_HUFFMAN_ENV_1_5DB, 48 F_HUFFMAN_ENV_1_5DB, 49 T_HUFFMAN_ENV_BAL_1_5DB, 50 F_HUFFMAN_ENV_BAL_1_5DB, 51 T_HUFFMAN_ENV_3_0DB, 52 F_HUFFMAN_ENV_3_0DB, 53 T_HUFFMAN_ENV_BAL_3_0DB, 54 F_HUFFMAN_ENV_BAL_3_0DB, 55 T_HUFFMAN_NOISE_3_0DB, 56 T_HUFFMAN_NOISE_BAL_3_0DB, 57}; 58 59/** 60 * bs_frame_class - frame class of current SBR frame (14496-3 sp04 p98) 61 */ 62enum { 63 FIXFIX, 64 FIXVAR, 65 VARFIX, 66 VARVAR, 67}; 68 69enum { 70 EXTENSION_ID_PS = 2, 71}; 72 73static VLC vlc_sbr[10]; 74static const int8_t vlc_sbr_lav[10] = 75 { 60, 60, 24, 24, 31, 31, 12, 12, 31, 12 }; 76static const DECLARE_ALIGNED(16, float, zero64)[64]; 77 78#define SBR_INIT_VLC_STATIC(num, size) \ 79 INIT_VLC_STATIC(&vlc_sbr[num], 9, sbr_tmp[num].table_size / sbr_tmp[num].elem_size, \ 80 sbr_tmp[num].sbr_bits , 1, 1, \ 81 sbr_tmp[num].sbr_codes, sbr_tmp[num].elem_size, sbr_tmp[num].elem_size, \ 82 size) 83 84#define SBR_VLC_ROW(name) \ 85 { name ## _codes, name ## _bits, sizeof(name ## _codes), sizeof(name ## _codes[0]) } 86 87av_cold void ff_aac_sbr_init(void) 88{ 89 int n; 90 static const struct { 91 const void *sbr_codes, *sbr_bits; 92 const unsigned int table_size, elem_size; 93 } sbr_tmp[] = { 94 SBR_VLC_ROW(t_huffman_env_1_5dB), 95 SBR_VLC_ROW(f_huffman_env_1_5dB), 96 SBR_VLC_ROW(t_huffman_env_bal_1_5dB), 97 SBR_VLC_ROW(f_huffman_env_bal_1_5dB), 98 SBR_VLC_ROW(t_huffman_env_3_0dB), 99 SBR_VLC_ROW(f_huffman_env_3_0dB), 100 SBR_VLC_ROW(t_huffman_env_bal_3_0dB), 101 SBR_VLC_ROW(f_huffman_env_bal_3_0dB), 102 SBR_VLC_ROW(t_huffman_noise_3_0dB), 103 SBR_VLC_ROW(t_huffman_noise_bal_3_0dB), 104 }; 105 106 // SBR VLC table initialization 107 SBR_INIT_VLC_STATIC(0, 1098); 108 SBR_INIT_VLC_STATIC(1, 1092); 109 SBR_INIT_VLC_STATIC(2, 768); 110 SBR_INIT_VLC_STATIC(3, 1026); 111 SBR_INIT_VLC_STATIC(4, 1058); 112 SBR_INIT_VLC_STATIC(5, 1052); 113 SBR_INIT_VLC_STATIC(6, 544); 114 SBR_INIT_VLC_STATIC(7, 544); 115 SBR_INIT_VLC_STATIC(8, 592); 116 SBR_INIT_VLC_STATIC(9, 512); 117 118 for (n = 1; n < 320; n++) 119 sbr_qmf_window_us[320 + n] = sbr_qmf_window_us[320 - n]; 120 sbr_qmf_window_us[384] = -sbr_qmf_window_us[384]; 121 sbr_qmf_window_us[512] = -sbr_qmf_window_us[512]; 122 123 for (n = 0; n < 320; n++) 124 sbr_qmf_window_ds[n] = sbr_qmf_window_us[2*n]; 125 126 ff_ps_init(); 127} 128 129av_cold void ff_aac_sbr_ctx_init(AACContext *ac, SpectralBandReplication *sbr) 130{ 131 float mdct_scale; 132 sbr->kx[0] = sbr->kx[1] = 32; //Typo in spec, kx' inits to 32 133 sbr->data[0].e_a[1] = sbr->data[1].e_a[1] = -1; 134 sbr->data[0].synthesis_filterbank_samples_offset = SBR_SYNTHESIS_BUF_SIZE - (1280 - 128); 135 sbr->data[1].synthesis_filterbank_samples_offset = SBR_SYNTHESIS_BUF_SIZE - (1280 - 128); 136 /* SBR requires samples to be scaled to +/-32768.0 to work correctly. 137 * mdct scale factors are adjusted to scale up from +/-1.0 at analysis 138 * and scale back down at synthesis. */ 139 mdct_scale = ac->avctx->sample_fmt == AV_SAMPLE_FMT_FLT ? 32768.0f : 1.0f; 140 ff_mdct_init(&sbr->mdct, 7, 1, 1.0 / (64 * mdct_scale)); 141 ff_mdct_init(&sbr->mdct_ana, 7, 1, -2.0 * mdct_scale); 142 ff_ps_ctx_init(&sbr->ps); 143} 144 145av_cold void ff_aac_sbr_ctx_close(SpectralBandReplication *sbr) 146{ 147 ff_mdct_end(&sbr->mdct); 148 ff_mdct_end(&sbr->mdct_ana); 149} 150 151static int qsort_comparison_function_int16(const void *a, const void *b) 152{ 153 return *(const int16_t *)a - *(const int16_t *)b; 154} 155 156static inline int in_table_int16(const int16_t *table, int last_el, int16_t needle) 157{ 158 int i; 159 for (i = 0; i <= last_el; i++) 160 if (table[i] == needle) 161 return 1; 162 return 0; 163} 164 165/// Limiter Frequency Band Table (14496-3 sp04 p198) 166static void sbr_make_f_tablelim(SpectralBandReplication *sbr) 167{ 168 int k; 169 if (sbr->bs_limiter_bands > 0) { 170 static const float bands_warped[3] = { 1.32715174233856803909f, //2^(0.49/1.2) 171 1.18509277094158210129f, //2^(0.49/2) 172 1.11987160404675912501f }; //2^(0.49/3) 173 const float lim_bands_per_octave_warped = bands_warped[sbr->bs_limiter_bands - 1]; 174 int16_t patch_borders[7]; 175 uint16_t *in = sbr->f_tablelim + 1, *out = sbr->f_tablelim; 176 177 patch_borders[0] = sbr->kx[1]; 178 for (k = 1; k <= sbr->num_patches; k++) 179 patch_borders[k] = patch_borders[k-1] + sbr->patch_num_subbands[k-1]; 180 181 memcpy(sbr->f_tablelim, sbr->f_tablelow, 182 (sbr->n[0] + 1) * sizeof(sbr->f_tablelow[0])); 183 if (sbr->num_patches > 1) 184 memcpy(sbr->f_tablelim + sbr->n[0] + 1, patch_borders + 1, 185 (sbr->num_patches - 1) * sizeof(patch_borders[0])); 186 187 qsort(sbr->f_tablelim, sbr->num_patches + sbr->n[0], 188 sizeof(sbr->f_tablelim[0]), 189 qsort_comparison_function_int16); 190 191 sbr->n_lim = sbr->n[0] + sbr->num_patches - 1; 192 while (out < sbr->f_tablelim + sbr->n_lim) { 193 if (*in >= *out * lim_bands_per_octave_warped) { 194 *++out = *in++; 195 } else if (*in == *out || 196 !in_table_int16(patch_borders, sbr->num_patches, *in)) { 197 in++; 198 sbr->n_lim--; 199 } else if (!in_table_int16(patch_borders, sbr->num_patches, *out)) { 200 *out = *in++; 201 sbr->n_lim--; 202 } else { 203 *++out = *in++; 204 } 205 } 206 } else { 207 sbr->f_tablelim[0] = sbr->f_tablelow[0]; 208 sbr->f_tablelim[1] = sbr->f_tablelow[sbr->n[0]]; 209 sbr->n_lim = 1; 210 } 211} 212 213static unsigned int read_sbr_header(SpectralBandReplication *sbr, GetBitContext *gb) 214{ 215 unsigned int cnt = get_bits_count(gb); 216 uint8_t bs_header_extra_1; 217 uint8_t bs_header_extra_2; 218 int old_bs_limiter_bands = sbr->bs_limiter_bands; 219 SpectrumParameters old_spectrum_params; 220 221 sbr->start = 1; 222 223 // Save last spectrum parameters variables to compare to new ones 224 memcpy(&old_spectrum_params, &sbr->spectrum_params, sizeof(SpectrumParameters)); 225 226 sbr->bs_amp_res_header = get_bits1(gb); 227 sbr->spectrum_params.bs_start_freq = get_bits(gb, 4); 228 sbr->spectrum_params.bs_stop_freq = get_bits(gb, 4); 229 sbr->spectrum_params.bs_xover_band = get_bits(gb, 3); 230 skip_bits(gb, 2); // bs_reserved 231 232 bs_header_extra_1 = get_bits1(gb); 233 bs_header_extra_2 = get_bits1(gb); 234 235 if (bs_header_extra_1) { 236 sbr->spectrum_params.bs_freq_scale = get_bits(gb, 2); 237 sbr->spectrum_params.bs_alter_scale = get_bits1(gb); 238 sbr->spectrum_params.bs_noise_bands = get_bits(gb, 2); 239 } else { 240 sbr->spectrum_params.bs_freq_scale = 2; 241 sbr->spectrum_params.bs_alter_scale = 1; 242 sbr->spectrum_params.bs_noise_bands = 2; 243 } 244 245 // Check if spectrum parameters changed 246 if (memcmp(&old_spectrum_params, &sbr->spectrum_params, sizeof(SpectrumParameters))) 247 sbr->reset = 1; 248 249 if (bs_header_extra_2) { 250 sbr->bs_limiter_bands = get_bits(gb, 2); 251 sbr->bs_limiter_gains = get_bits(gb, 2); 252 sbr->bs_interpol_freq = get_bits1(gb); 253 sbr->bs_smoothing_mode = get_bits1(gb); 254 } else { 255 sbr->bs_limiter_bands = 2; 256 sbr->bs_limiter_gains = 2; 257 sbr->bs_interpol_freq = 1; 258 sbr->bs_smoothing_mode = 1; 259 } 260 261 if (sbr->bs_limiter_bands != old_bs_limiter_bands && !sbr->reset) 262 sbr_make_f_tablelim(sbr); 263 264 return get_bits_count(gb) - cnt; 265} 266 267static int array_min_int16(const int16_t *array, int nel) 268{ 269 int i, min = array[0]; 270 for (i = 1; i < nel; i++) 271 min = FFMIN(array[i], min); 272 return min; 273} 274 275static void make_bands(int16_t* bands, int start, int stop, int num_bands) 276{ 277 int k, previous, present; 278 float base, prod; 279 280 base = powf((float)stop / start, 1.0f / num_bands); 281 prod = start; 282 previous = start; 283 284 for (k = 0; k < num_bands-1; k++) { 285 prod *= base; 286 present = lrintf(prod); 287 bands[k] = present - previous; 288 previous = present; 289 } 290 bands[num_bands-1] = stop - previous; 291} 292 293static int check_n_master(AVCodecContext *avctx, int n_master, int bs_xover_band) 294{ 295 // Requirements (14496-3 sp04 p205) 296 if (n_master <= 0) { 297 av_log(avctx, AV_LOG_ERROR, "Invalid n_master: %d\n", n_master); 298 return -1; 299 } 300 if (bs_xover_band >= n_master) { 301 av_log(avctx, AV_LOG_ERROR, 302 "Invalid bitstream, crossover band index beyond array bounds: %d\n", 303 bs_xover_band); 304 return -1; 305 } 306 return 0; 307} 308 309/// Master Frequency Band Table (14496-3 sp04 p194) 310static int sbr_make_f_master(AACContext *ac, SpectralBandReplication *sbr, 311 SpectrumParameters *spectrum) 312{ 313 unsigned int temp, max_qmf_subbands; 314 unsigned int start_min, stop_min; 315 int k; 316 const int8_t *sbr_offset_ptr; 317 int16_t stop_dk[13]; 318 319 if (sbr->sample_rate < 32000) { 320 temp = 3000; 321 } else if (sbr->sample_rate < 64000) { 322 temp = 4000; 323 } else 324 temp = 5000; 325 326 start_min = ((temp << 7) + (sbr->sample_rate >> 1)) / sbr->sample_rate; 327 stop_min = ((temp << 8) + (sbr->sample_rate >> 1)) / sbr->sample_rate; 328 329 switch (sbr->sample_rate) { 330 case 16000: 331 sbr_offset_ptr = sbr_offset[0]; 332 break; 333 case 22050: 334 sbr_offset_ptr = sbr_offset[1]; 335 break; 336 case 24000: 337 sbr_offset_ptr = sbr_offset[2]; 338 break; 339 case 32000: 340 sbr_offset_ptr = sbr_offset[3]; 341 break; 342 case 44100: case 48000: case 64000: 343 sbr_offset_ptr = sbr_offset[4]; 344 break; 345 case 88200: case 96000: case 128000: case 176400: case 192000: 346 sbr_offset_ptr = sbr_offset[5]; 347 break; 348 default: 349 av_log(ac->avctx, AV_LOG_ERROR, 350 "Unsupported sample rate for SBR: %d\n", sbr->sample_rate); 351 return -1; 352 } 353 354 sbr->k[0] = start_min + sbr_offset_ptr[spectrum->bs_start_freq]; 355 356 if (spectrum->bs_stop_freq < 14) { 357 sbr->k[2] = stop_min; 358 make_bands(stop_dk, stop_min, 64, 13); 359 qsort(stop_dk, 13, sizeof(stop_dk[0]), qsort_comparison_function_int16); 360 for (k = 0; k < spectrum->bs_stop_freq; k++) 361 sbr->k[2] += stop_dk[k]; 362 } else if (spectrum->bs_stop_freq == 14) { 363 sbr->k[2] = 2*sbr->k[0]; 364 } else if (spectrum->bs_stop_freq == 15) { 365 sbr->k[2] = 3*sbr->k[0]; 366 } else { 367 av_log(ac->avctx, AV_LOG_ERROR, 368 "Invalid bs_stop_freq: %d\n", spectrum->bs_stop_freq); 369 return -1; 370 } 371 sbr->k[2] = FFMIN(64, sbr->k[2]); 372 373 // Requirements (14496-3 sp04 p205) 374 if (sbr->sample_rate <= 32000) { 375 max_qmf_subbands = 48; 376 } else if (sbr->sample_rate == 44100) { 377 max_qmf_subbands = 35; 378 } else if (sbr->sample_rate >= 48000) 379 max_qmf_subbands = 32; 380 381 if (sbr->k[2] - sbr->k[0] > max_qmf_subbands) { 382 av_log(ac->avctx, AV_LOG_ERROR, 383 "Invalid bitstream, too many QMF subbands: %d\n", sbr->k[2] - sbr->k[0]); 384 return -1; 385 } 386 387 if (!spectrum->bs_freq_scale) { 388 int dk, k2diff; 389 390 dk = spectrum->bs_alter_scale + 1; 391 sbr->n_master = ((sbr->k[2] - sbr->k[0] + (dk&2)) >> dk) << 1; 392 if (check_n_master(ac->avctx, sbr->n_master, sbr->spectrum_params.bs_xover_band)) 393 return -1; 394 395 for (k = 1; k <= sbr->n_master; k++) 396 sbr->f_master[k] = dk; 397 398 k2diff = sbr->k[2] - sbr->k[0] - sbr->n_master * dk; 399 if (k2diff < 0) { 400 sbr->f_master[1]--; 401 sbr->f_master[2]-= (k2diff < -1); 402 } else if (k2diff) { 403 sbr->f_master[sbr->n_master]++; 404 } 405 406 sbr->f_master[0] = sbr->k[0]; 407 for (k = 1; k <= sbr->n_master; k++) 408 sbr->f_master[k] += sbr->f_master[k - 1]; 409 410 } else { 411 int half_bands = 7 - spectrum->bs_freq_scale; // bs_freq_scale = {1,2,3} 412 int two_regions, num_bands_0; 413 int vdk0_max, vdk1_min; 414 int16_t vk0[49]; 415 416 if (49 * sbr->k[2] > 110 * sbr->k[0]) { 417 two_regions = 1; 418 sbr->k[1] = 2 * sbr->k[0]; 419 } else { 420 two_regions = 0; 421 sbr->k[1] = sbr->k[2]; 422 } 423 424 num_bands_0 = lrintf(half_bands * log2f(sbr->k[1] / (float)sbr->k[0])) * 2; 425 426 if (num_bands_0 <= 0) { // Requirements (14496-3 sp04 p205) 427 av_log(ac->avctx, AV_LOG_ERROR, "Invalid num_bands_0: %d\n", num_bands_0); 428 return -1; 429 } 430 431 vk0[0] = 0; 432 433 make_bands(vk0+1, sbr->k[0], sbr->k[1], num_bands_0); 434 435 qsort(vk0 + 1, num_bands_0, sizeof(vk0[1]), qsort_comparison_function_int16); 436 vdk0_max = vk0[num_bands_0]; 437 438 vk0[0] = sbr->k[0]; 439 for (k = 1; k <= num_bands_0; k++) { 440 if (vk0[k] <= 0) { // Requirements (14496-3 sp04 p205) 441 av_log(ac->avctx, AV_LOG_ERROR, "Invalid vDk0[%d]: %d\n", k, vk0[k]); 442 return -1; 443 } 444 vk0[k] += vk0[k-1]; 445 } 446 447 if (two_regions) { 448 int16_t vk1[49]; 449 float invwarp = spectrum->bs_alter_scale ? 0.76923076923076923077f 450 : 1.0f; // bs_alter_scale = {0,1} 451 int num_bands_1 = lrintf(half_bands * invwarp * 452 log2f(sbr->k[2] / (float)sbr->k[1])) * 2; 453 454 make_bands(vk1+1, sbr->k[1], sbr->k[2], num_bands_1); 455 456 vdk1_min = array_min_int16(vk1 + 1, num_bands_1); 457 458 if (vdk1_min < vdk0_max) { 459 int change; 460 qsort(vk1 + 1, num_bands_1, sizeof(vk1[1]), qsort_comparison_function_int16); 461 change = FFMIN(vdk0_max - vk1[1], (vk1[num_bands_1] - vk1[1]) >> 1); 462 vk1[1] += change; 463 vk1[num_bands_1] -= change; 464 } 465 466 qsort(vk1 + 1, num_bands_1, sizeof(vk1[1]), qsort_comparison_function_int16); 467 468 vk1[0] = sbr->k[1]; 469 for (k = 1; k <= num_bands_1; k++) { 470 if (vk1[k] <= 0) { // Requirements (14496-3 sp04 p205) 471 av_log(ac->avctx, AV_LOG_ERROR, "Invalid vDk1[%d]: %d\n", k, vk1[k]); 472 return -1; 473 } 474 vk1[k] += vk1[k-1]; 475 } 476 477 sbr->n_master = num_bands_0 + num_bands_1; 478 if (check_n_master(ac->avctx, sbr->n_master, sbr->spectrum_params.bs_xover_band)) 479 return -1; 480 memcpy(&sbr->f_master[0], vk0, 481 (num_bands_0 + 1) * sizeof(sbr->f_master[0])); 482 memcpy(&sbr->f_master[num_bands_0 + 1], vk1 + 1, 483 num_bands_1 * sizeof(sbr->f_master[0])); 484 485 } else { 486 sbr->n_master = num_bands_0; 487 if (check_n_master(ac->avctx, sbr->n_master, sbr->spectrum_params.bs_xover_band)) 488 return -1; 489 memcpy(sbr->f_master, vk0, (num_bands_0 + 1) * sizeof(sbr->f_master[0])); 490 } 491 } 492 493 return 0; 494} 495 496/// High Frequency Generation - Patch Construction (14496-3 sp04 p216 fig. 4.46) 497static int sbr_hf_calc_npatches(AACContext *ac, SpectralBandReplication *sbr) 498{ 499 int i, k, sb = 0; 500 int msb = sbr->k[0]; 501 int usb = sbr->kx[1]; 502 int goal_sb = ((1000 << 11) + (sbr->sample_rate >> 1)) / sbr->sample_rate; 503 504 sbr->num_patches = 0; 505 506 if (goal_sb < sbr->kx[1] + sbr->m[1]) { 507 for (k = 0; sbr->f_master[k] < goal_sb; k++) ; 508 } else 509 k = sbr->n_master; 510 511 do { 512 int odd = 0; 513 for (i = k; i == k || sb > (sbr->k[0] - 1 + msb - odd); i--) { 514 sb = sbr->f_master[i]; 515 odd = (sb + sbr->k[0]) & 1; 516 } 517 518 // Requirements (14496-3 sp04 p205) sets the maximum number of patches to 5. 519 // After this check the final number of patches can still be six which is 520 // illegal however the Coding Technologies decoder check stream has a final 521 // count of 6 patches 522 if (sbr->num_patches > 5) { 523 av_log(ac->avctx, AV_LOG_ERROR, "Too many patches: %d\n", sbr->num_patches); 524 return -1; 525 } 526 527 sbr->patch_num_subbands[sbr->num_patches] = FFMAX(sb - usb, 0); 528 sbr->patch_start_subband[sbr->num_patches] = sbr->k[0] - odd - sbr->patch_num_subbands[sbr->num_patches]; 529 530 if (sbr->patch_num_subbands[sbr->num_patches] > 0) { 531 usb = sb; 532 msb = sb; 533 sbr->num_patches++; 534 } else 535 msb = sbr->kx[1]; 536 537 if (sbr->f_master[k] - sb < 3) 538 k = sbr->n_master; 539 } while (sb != sbr->kx[1] + sbr->m[1]); 540 541 if (sbr->patch_num_subbands[sbr->num_patches-1] < 3 && sbr->num_patches > 1) 542 sbr->num_patches--; 543 544 return 0; 545} 546 547/// Derived Frequency Band Tables (14496-3 sp04 p197) 548static int sbr_make_f_derived(AACContext *ac, SpectralBandReplication *sbr) 549{ 550 int k, temp; 551 552 sbr->n[1] = sbr->n_master - sbr->spectrum_params.bs_xover_band; 553 sbr->n[0] = (sbr->n[1] + 1) >> 1; 554 555 memcpy(sbr->f_tablehigh, &sbr->f_master[sbr->spectrum_params.bs_xover_band], 556 (sbr->n[1] + 1) * sizeof(sbr->f_master[0])); 557 sbr->m[1] = sbr->f_tablehigh[sbr->n[1]] - sbr->f_tablehigh[0]; 558 sbr->kx[1] = sbr->f_tablehigh[0]; 559 560 // Requirements (14496-3 sp04 p205) 561 if (sbr->kx[1] + sbr->m[1] > 64) { 562 av_log(ac->avctx, AV_LOG_ERROR, 563 "Stop frequency border too high: %d\n", sbr->kx[1] + sbr->m[1]); 564 return -1; 565 } 566 if (sbr->kx[1] > 32) { 567 av_log(ac->avctx, AV_LOG_ERROR, "Start frequency border too high: %d\n", sbr->kx[1]); 568 return -1; 569 } 570 571 sbr->f_tablelow[0] = sbr->f_tablehigh[0]; 572 temp = sbr->n[1] & 1; 573 for (k = 1; k <= sbr->n[0]; k++) 574 sbr->f_tablelow[k] = sbr->f_tablehigh[2 * k - temp]; 575 576 sbr->n_q = FFMAX(1, lrintf(sbr->spectrum_params.bs_noise_bands * 577 log2f(sbr->k[2] / (float)sbr->kx[1]))); // 0 <= bs_noise_bands <= 3 578 if (sbr->n_q > 5) { 579 av_log(ac->avctx, AV_LOG_ERROR, "Too many noise floor scale factors: %d\n", sbr->n_q); 580 return -1; 581 } 582 583 sbr->f_tablenoise[0] = sbr->f_tablelow[0]; 584 temp = 0; 585 for (k = 1; k <= sbr->n_q; k++) { 586 temp += (sbr->n[0] - temp) / (sbr->n_q + 1 - k); 587 sbr->f_tablenoise[k] = sbr->f_tablelow[temp]; 588 } 589 590 if (sbr_hf_calc_npatches(ac, sbr) < 0) 591 return -1; 592 593 sbr_make_f_tablelim(sbr); 594 595 sbr->data[0].f_indexnoise = 0; 596 sbr->data[1].f_indexnoise = 0; 597 598 return 0; 599} 600 601static av_always_inline void get_bits1_vector(GetBitContext *gb, uint8_t *vec, 602 int elements) 603{ 604 int i; 605 for (i = 0; i < elements; i++) { 606 vec[i] = get_bits1(gb); 607 } 608} 609 610/** ceil(log2(index+1)) */ 611static const int8_t ceil_log2[] = { 612 0, 1, 2, 2, 3, 3, 613}; 614 615static int read_sbr_grid(AACContext *ac, SpectralBandReplication *sbr, 616 GetBitContext *gb, SBRData *ch_data) 617{ 618 int i; 619 unsigned bs_pointer = 0; 620 // frameLengthFlag ? 15 : 16; 960 sample length frames unsupported; this value is numTimeSlots 621 int abs_bord_trail = 16; 622 int num_rel_lead, num_rel_trail; 623 unsigned bs_num_env_old = ch_data->bs_num_env; 624 625 ch_data->bs_freq_res[0] = ch_data->bs_freq_res[ch_data->bs_num_env]; 626 ch_data->bs_amp_res = sbr->bs_amp_res_header; 627 ch_data->t_env_num_env_old = ch_data->t_env[bs_num_env_old]; 628 629 switch (ch_data->bs_frame_class = get_bits(gb, 2)) { 630 case FIXFIX: 631 ch_data->bs_num_env = 1 << get_bits(gb, 2); 632 num_rel_lead = ch_data->bs_num_env - 1; 633 if (ch_data->bs_num_env == 1) 634 ch_data->bs_amp_res = 0; 635 636 if (ch_data->bs_num_env > 4) { 637 av_log(ac->avctx, AV_LOG_ERROR, 638 "Invalid bitstream, too many SBR envelopes in FIXFIX type SBR frame: %d\n", 639 ch_data->bs_num_env); 640 return -1; 641 } 642 643 ch_data->t_env[0] = 0; 644 ch_data->t_env[ch_data->bs_num_env] = abs_bord_trail; 645 646 abs_bord_trail = (abs_bord_trail + (ch_data->bs_num_env >> 1)) / 647 ch_data->bs_num_env; 648 for (i = 0; i < num_rel_lead; i++) 649 ch_data->t_env[i + 1] = ch_data->t_env[i] + abs_bord_trail; 650 651 ch_data->bs_freq_res[1] = get_bits1(gb); 652 for (i = 1; i < ch_data->bs_num_env; i++) 653 ch_data->bs_freq_res[i + 1] = ch_data->bs_freq_res[1]; 654 break; 655 case FIXVAR: 656 abs_bord_trail += get_bits(gb, 2); 657 num_rel_trail = get_bits(gb, 2); 658 ch_data->bs_num_env = num_rel_trail + 1; 659 ch_data->t_env[0] = 0; 660 ch_data->t_env[ch_data->bs_num_env] = abs_bord_trail; 661 662 for (i = 0; i < num_rel_trail; i++) 663 ch_data->t_env[ch_data->bs_num_env - 1 - i] = 664 ch_data->t_env[ch_data->bs_num_env - i] - 2 * get_bits(gb, 2) - 2; 665 666 bs_pointer = get_bits(gb, ceil_log2[ch_data->bs_num_env]); 667 668 for (i = 0; i < ch_data->bs_num_env; i++) 669 ch_data->bs_freq_res[ch_data->bs_num_env - i] = get_bits1(gb); 670 break; 671 case VARFIX: 672 ch_data->t_env[0] = get_bits(gb, 2); 673 num_rel_lead = get_bits(gb, 2); 674 ch_data->bs_num_env = num_rel_lead + 1; 675 ch_data->t_env[ch_data->bs_num_env] = abs_bord_trail; 676 677 for (i = 0; i < num_rel_lead; i++) 678 ch_data->t_env[i + 1] = ch_data->t_env[i] + 2 * get_bits(gb, 2) + 2; 679 680 bs_pointer = get_bits(gb, ceil_log2[ch_data->bs_num_env]); 681 682 get_bits1_vector(gb, ch_data->bs_freq_res + 1, ch_data->bs_num_env); 683 break; 684 case VARVAR: 685 ch_data->t_env[0] = get_bits(gb, 2); 686 abs_bord_trail += get_bits(gb, 2); 687 num_rel_lead = get_bits(gb, 2); 688 num_rel_trail = get_bits(gb, 2); 689 ch_data->bs_num_env = num_rel_lead + num_rel_trail + 1; 690 691 if (ch_data->bs_num_env > 5) { 692 av_log(ac->avctx, AV_LOG_ERROR, 693 "Invalid bitstream, too many SBR envelopes in VARVAR type SBR frame: %d\n", 694 ch_data->bs_num_env); 695 return -1; 696 } 697 698 ch_data->t_env[ch_data->bs_num_env] = abs_bord_trail; 699 700 for (i = 0; i < num_rel_lead; i++) 701 ch_data->t_env[i + 1] = ch_data->t_env[i] + 2 * get_bits(gb, 2) + 2; 702 for (i = 0; i < num_rel_trail; i++) 703 ch_data->t_env[ch_data->bs_num_env - 1 - i] = 704 ch_data->t_env[ch_data->bs_num_env - i] - 2 * get_bits(gb, 2) - 2; 705 706 bs_pointer = get_bits(gb, ceil_log2[ch_data->bs_num_env]); 707 708 get_bits1_vector(gb, ch_data->bs_freq_res + 1, ch_data->bs_num_env); 709 break; 710 } 711 712 if (bs_pointer > ch_data->bs_num_env + 1) { 713 av_log(ac->avctx, AV_LOG_ERROR, 714 "Invalid bitstream, bs_pointer points to a middle noise border outside the time borders table: %d\n", 715 bs_pointer); 716 return -1; 717 } 718 719 for (i = 1; i <= ch_data->bs_num_env; i++) { 720 if (ch_data->t_env[i-1] > ch_data->t_env[i]) { 721 av_log(ac->avctx, AV_LOG_ERROR, "Non monotone time borders\n"); 722 return -1; 723 } 724 } 725 726 ch_data->bs_num_noise = (ch_data->bs_num_env > 1) + 1; 727 728 ch_data->t_q[0] = ch_data->t_env[0]; 729 ch_data->t_q[ch_data->bs_num_noise] = ch_data->t_env[ch_data->bs_num_env]; 730 if (ch_data->bs_num_noise > 1) { 731 unsigned int idx; 732 if (ch_data->bs_frame_class == FIXFIX) { 733 idx = ch_data->bs_num_env >> 1; 734 } else if (ch_data->bs_frame_class & 1) { // FIXVAR or VARVAR 735 idx = ch_data->bs_num_env - FFMAX(bs_pointer - 1, 1); 736 } else { // VARFIX 737 if (!bs_pointer) 738 idx = 1; 739 else if (bs_pointer == 1) 740 idx = ch_data->bs_num_env - 1; 741 else // bs_pointer > 1 742 idx = bs_pointer - 1; 743 } 744 ch_data->t_q[1] = ch_data->t_env[idx]; 745 } 746 747 ch_data->e_a[0] = -(ch_data->e_a[1] != bs_num_env_old); // l_APrev 748 ch_data->e_a[1] = -1; 749 if ((ch_data->bs_frame_class & 1) && bs_pointer) { // FIXVAR or VARVAR and bs_pointer != 0 750 ch_data->e_a[1] = ch_data->bs_num_env + 1 - bs_pointer; 751 } else if ((ch_data->bs_frame_class == 2) && (bs_pointer > 1)) // VARFIX and bs_pointer > 1 752 ch_data->e_a[1] = bs_pointer - 1; 753 754 return 0; 755} 756 757static void copy_sbr_grid(SBRData *dst, const SBRData *src) { 758 //These variables are saved from the previous frame rather than copied 759 dst->bs_freq_res[0] = dst->bs_freq_res[dst->bs_num_env]; 760 dst->t_env_num_env_old = dst->t_env[dst->bs_num_env]; 761 dst->e_a[0] = -(dst->e_a[1] != dst->bs_num_env); 762 763 //These variables are read from the bitstream and therefore copied 764 memcpy(dst->bs_freq_res+1, src->bs_freq_res+1, sizeof(dst->bs_freq_res)-sizeof(*dst->bs_freq_res)); 765 memcpy(dst->t_env, src->t_env, sizeof(dst->t_env)); 766 memcpy(dst->t_q, src->t_q, sizeof(dst->t_q)); 767 dst->bs_num_env = src->bs_num_env; 768 dst->bs_amp_res = src->bs_amp_res; 769 dst->bs_num_noise = src->bs_num_noise; 770 dst->bs_frame_class = src->bs_frame_class; 771 dst->e_a[1] = src->e_a[1]; 772} 773 774/// Read how the envelope and noise floor data is delta coded 775static void read_sbr_dtdf(SpectralBandReplication *sbr, GetBitContext *gb, 776 SBRData *ch_data) 777{ 778 get_bits1_vector(gb, ch_data->bs_df_env, ch_data->bs_num_env); 779 get_bits1_vector(gb, ch_data->bs_df_noise, ch_data->bs_num_noise); 780} 781 782/// Read inverse filtering data 783static void read_sbr_invf(SpectralBandReplication *sbr, GetBitContext *gb, 784 SBRData *ch_data) 785{ 786 int i; 787 788 memcpy(ch_data->bs_invf_mode[1], ch_data->bs_invf_mode[0], 5 * sizeof(uint8_t)); 789 for (i = 0; i < sbr->n_q; i++) 790 ch_data->bs_invf_mode[0][i] = get_bits(gb, 2); 791} 792 793static void read_sbr_envelope(SpectralBandReplication *sbr, GetBitContext *gb, 794 SBRData *ch_data, int ch) 795{ 796 int bits; 797 int i, j, k; 798 VLC_TYPE (*t_huff)[2], (*f_huff)[2]; 799 int t_lav, f_lav; 800 const int delta = (ch == 1 && sbr->bs_coupling == 1) + 1; 801 const int odd = sbr->n[1] & 1; 802 803 if (sbr->bs_coupling && ch) { 804 if (ch_data->bs_amp_res) { 805 bits = 5; 806 t_huff = vlc_sbr[T_HUFFMAN_ENV_BAL_3_0DB].table; 807 t_lav = vlc_sbr_lav[T_HUFFMAN_ENV_BAL_3_0DB]; 808 f_huff = vlc_sbr[F_HUFFMAN_ENV_BAL_3_0DB].table; 809 f_lav = vlc_sbr_lav[F_HUFFMAN_ENV_BAL_3_0DB]; 810 } else { 811 bits = 6; 812 t_huff = vlc_sbr[T_HUFFMAN_ENV_BAL_1_5DB].table; 813 t_lav = vlc_sbr_lav[T_HUFFMAN_ENV_BAL_1_5DB]; 814 f_huff = vlc_sbr[F_HUFFMAN_ENV_BAL_1_5DB].table; 815 f_lav = vlc_sbr_lav[F_HUFFMAN_ENV_BAL_1_5DB]; 816 } 817 } else { 818 if (ch_data->bs_amp_res) { 819 bits = 6; 820 t_huff = vlc_sbr[T_HUFFMAN_ENV_3_0DB].table; 821 t_lav = vlc_sbr_lav[T_HUFFMAN_ENV_3_0DB]; 822 f_huff = vlc_sbr[F_HUFFMAN_ENV_3_0DB].table; 823 f_lav = vlc_sbr_lav[F_HUFFMAN_ENV_3_0DB]; 824 } else { 825 bits = 7; 826 t_huff = vlc_sbr[T_HUFFMAN_ENV_1_5DB].table; 827 t_lav = vlc_sbr_lav[T_HUFFMAN_ENV_1_5DB]; 828 f_huff = vlc_sbr[F_HUFFMAN_ENV_1_5DB].table; 829 f_lav = vlc_sbr_lav[F_HUFFMAN_ENV_1_5DB]; 830 } 831 } 832 833 for (i = 0; i < ch_data->bs_num_env; i++) { 834 if (ch_data->bs_df_env[i]) { 835 // bs_freq_res[0] == bs_freq_res[bs_num_env] from prev frame 836 if (ch_data->bs_freq_res[i + 1] == ch_data->bs_freq_res[i]) { 837 for (j = 0; j < sbr->n[ch_data->bs_freq_res[i + 1]]; j++) 838 ch_data->env_facs[i + 1][j] = ch_data->env_facs[i][j] + delta * (get_vlc2(gb, t_huff, 9, 3) - t_lav); 839 } else if (ch_data->bs_freq_res[i + 1]) { 840 for (j = 0; j < sbr->n[ch_data->bs_freq_res[i + 1]]; j++) { 841 k = (j + odd) >> 1; // find k such that f_tablelow[k] <= f_tablehigh[j] < f_tablelow[k + 1] 842 ch_data->env_facs[i + 1][j] = ch_data->env_facs[i][k] + delta * (get_vlc2(gb, t_huff, 9, 3) - t_lav); 843 } 844 } else { 845 for (j = 0; j < sbr->n[ch_data->bs_freq_res[i + 1]]; j++) { 846 k = j ? 2*j - odd : 0; // find k such that f_tablehigh[k] == f_tablelow[j] 847 ch_data->env_facs[i + 1][j] = ch_data->env_facs[i][k] + delta * (get_vlc2(gb, t_huff, 9, 3) - t_lav); 848 } 849 } 850 } else { 851 ch_data->env_facs[i + 1][0] = delta * get_bits(gb, bits); // bs_env_start_value_balance 852 for (j = 1; j < sbr->n[ch_data->bs_freq_res[i + 1]]; j++) 853 ch_data->env_facs[i + 1][j] = ch_data->env_facs[i + 1][j - 1] + delta * (get_vlc2(gb, f_huff, 9, 3) - f_lav); 854 } 855 } 856 857 //assign 0th elements of env_facs from last elements 858 memcpy(ch_data->env_facs[0], ch_data->env_facs[ch_data->bs_num_env], 859 sizeof(ch_data->env_facs[0])); 860} 861 862static void read_sbr_noise(SpectralBandReplication *sbr, GetBitContext *gb, 863 SBRData *ch_data, int ch) 864{ 865 int i, j; 866 VLC_TYPE (*t_huff)[2], (*f_huff)[2]; 867 int t_lav, f_lav; 868 int delta = (ch == 1 && sbr->bs_coupling == 1) + 1; 869 870 if (sbr->bs_coupling && ch) { 871 t_huff = vlc_sbr[T_HUFFMAN_NOISE_BAL_3_0DB].table; 872 t_lav = vlc_sbr_lav[T_HUFFMAN_NOISE_BAL_3_0DB]; 873 f_huff = vlc_sbr[F_HUFFMAN_ENV_BAL_3_0DB].table; 874 f_lav = vlc_sbr_lav[F_HUFFMAN_ENV_BAL_3_0DB]; 875 } else { 876 t_huff = vlc_sbr[T_HUFFMAN_NOISE_3_0DB].table; 877 t_lav = vlc_sbr_lav[T_HUFFMAN_NOISE_3_0DB]; 878 f_huff = vlc_sbr[F_HUFFMAN_ENV_3_0DB].table; 879 f_lav = vlc_sbr_lav[F_HUFFMAN_ENV_3_0DB]; 880 } 881 882 for (i = 0; i < ch_data->bs_num_noise; i++) { 883 if (ch_data->bs_df_noise[i]) { 884 for (j = 0; j < sbr->n_q; j++) 885 ch_data->noise_facs[i + 1][j] = ch_data->noise_facs[i][j] + delta * (get_vlc2(gb, t_huff, 9, 2) - t_lav); 886 } else { 887 ch_data->noise_facs[i + 1][0] = delta * get_bits(gb, 5); // bs_noise_start_value_balance or bs_noise_start_value_level 888 for (j = 1; j < sbr->n_q; j++) 889 ch_data->noise_facs[i + 1][j] = ch_data->noise_facs[i + 1][j - 1] + delta * (get_vlc2(gb, f_huff, 9, 3) - f_lav); 890 } 891 } 892 893 //assign 0th elements of noise_facs from last elements 894 memcpy(ch_data->noise_facs[0], ch_data->noise_facs[ch_data->bs_num_noise], 895 sizeof(ch_data->noise_facs[0])); 896} 897 898static void read_sbr_extension(AACContext *ac, SpectralBandReplication *sbr, 899 GetBitContext *gb, 900 int bs_extension_id, int *num_bits_left) 901{ 902 switch (bs_extension_id) { 903 case EXTENSION_ID_PS: 904 if (!ac->m4ac.ps) { 905 av_log(ac->avctx, AV_LOG_ERROR, "Parametric Stereo signaled to be not-present but was found in the bitstream.\n"); 906 skip_bits_long(gb, *num_bits_left); // bs_fill_bits 907 *num_bits_left = 0; 908 } else { 909#if 1 910 *num_bits_left -= ff_ps_read_data(ac->avctx, gb, &sbr->ps, *num_bits_left); 911#else 912 av_log_missing_feature(ac->avctx, "Parametric Stereo is", 0); 913 skip_bits_long(gb, *num_bits_left); // bs_fill_bits 914 *num_bits_left = 0; 915#endif 916 } 917 break; 918 default: 919 av_log_missing_feature(ac->avctx, "Reserved SBR extensions are", 1); 920 skip_bits_long(gb, *num_bits_left); // bs_fill_bits 921 *num_bits_left = 0; 922 break; 923 } 924} 925 926static int read_sbr_single_channel_element(AACContext *ac, 927 SpectralBandReplication *sbr, 928 GetBitContext *gb) 929{ 930 if (get_bits1(gb)) // bs_data_extra 931 skip_bits(gb, 4); // bs_reserved 932 933 if (read_sbr_grid(ac, sbr, gb, &sbr->data[0])) 934 return -1; 935 read_sbr_dtdf(sbr, gb, &sbr->data[0]); 936 read_sbr_invf(sbr, gb, &sbr->data[0]); 937 read_sbr_envelope(sbr, gb, &sbr->data[0], 0); 938 read_sbr_noise(sbr, gb, &sbr->data[0], 0); 939 940 if ((sbr->data[0].bs_add_harmonic_flag = get_bits1(gb))) 941 get_bits1_vector(gb, sbr->data[0].bs_add_harmonic, sbr->n[1]); 942 943 return 0; 944} 945 946static int read_sbr_channel_pair_element(AACContext *ac, 947 SpectralBandReplication *sbr, 948 GetBitContext *gb) 949{ 950 if (get_bits1(gb)) // bs_data_extra 951 skip_bits(gb, 8); // bs_reserved 952 953 if ((sbr->bs_coupling = get_bits1(gb))) { 954 if (read_sbr_grid(ac, sbr, gb, &sbr->data[0])) 955 return -1; 956 copy_sbr_grid(&sbr->data[1], &sbr->data[0]); 957 read_sbr_dtdf(sbr, gb, &sbr->data[0]); 958 read_sbr_dtdf(sbr, gb, &sbr->data[1]); 959 read_sbr_invf(sbr, gb, &sbr->data[0]); 960 memcpy(sbr->data[1].bs_invf_mode[1], sbr->data[1].bs_invf_mode[0], sizeof(sbr->data[1].bs_invf_mode[0])); 961 memcpy(sbr->data[1].bs_invf_mode[0], sbr->data[0].bs_invf_mode[0], sizeof(sbr->data[1].bs_invf_mode[0])); 962 read_sbr_envelope(sbr, gb, &sbr->data[0], 0); 963 read_sbr_noise(sbr, gb, &sbr->data[0], 0); 964 read_sbr_envelope(sbr, gb, &sbr->data[1], 1); 965 read_sbr_noise(sbr, gb, &sbr->data[1], 1); 966 } else { 967 if (read_sbr_grid(ac, sbr, gb, &sbr->data[0]) || 968 read_sbr_grid(ac, sbr, gb, &sbr->data[1])) 969 return -1; 970 read_sbr_dtdf(sbr, gb, &sbr->data[0]); 971 read_sbr_dtdf(sbr, gb, &sbr->data[1]); 972 read_sbr_invf(sbr, gb, &sbr->data[0]); 973 read_sbr_invf(sbr, gb, &sbr->data[1]); 974 read_sbr_envelope(sbr, gb, &sbr->data[0], 0); 975 read_sbr_envelope(sbr, gb, &sbr->data[1], 1); 976 read_sbr_noise(sbr, gb, &sbr->data[0], 0); 977 read_sbr_noise(sbr, gb, &sbr->data[1], 1); 978 } 979 980 if ((sbr->data[0].bs_add_harmonic_flag = get_bits1(gb))) 981 get_bits1_vector(gb, sbr->data[0].bs_add_harmonic, sbr->n[1]); 982 if ((sbr->data[1].bs_add_harmonic_flag = get_bits1(gb))) 983 get_bits1_vector(gb, sbr->data[1].bs_add_harmonic, sbr->n[1]); 984 985 return 0; 986} 987 988static unsigned int read_sbr_data(AACContext *ac, SpectralBandReplication *sbr, 989 GetBitContext *gb, int id_aac) 990{ 991 unsigned int cnt = get_bits_count(gb); 992 993 if (id_aac == TYPE_SCE || id_aac == TYPE_CCE) { 994 if (read_sbr_single_channel_element(ac, sbr, gb)) { 995 sbr->start = 0; 996 return get_bits_count(gb) - cnt; 997 } 998 } else if (id_aac == TYPE_CPE) { 999 if (read_sbr_channel_pair_element(ac, sbr, gb)) { 1000 sbr->start = 0; 1001 return get_bits_count(gb) - cnt; 1002 } 1003 } else { 1004 av_log(ac->avctx, AV_LOG_ERROR, 1005 "Invalid bitstream - cannot apply SBR to element type %d\n", id_aac); 1006 sbr->start = 0; 1007 return get_bits_count(gb) - cnt; 1008 } 1009 if (get_bits1(gb)) { // bs_extended_data 1010 int num_bits_left = get_bits(gb, 4); // bs_extension_size 1011 if (num_bits_left == 15) 1012 num_bits_left += get_bits(gb, 8); // bs_esc_count 1013 1014 num_bits_left <<= 3; 1015 while (num_bits_left > 7) { 1016 num_bits_left -= 2; 1017 read_sbr_extension(ac, sbr, gb, get_bits(gb, 2), &num_bits_left); // bs_extension_id 1018 } 1019 if (num_bits_left < 0) { 1020 av_log(ac->avctx, AV_LOG_ERROR, "SBR Extension over read.\n"); 1021 } 1022 if (num_bits_left > 0) 1023 skip_bits(gb, num_bits_left); 1024 } 1025 1026 return get_bits_count(gb) - cnt; 1027} 1028 1029static void sbr_reset(AACContext *ac, SpectralBandReplication *sbr) 1030{ 1031 int err; 1032 err = sbr_make_f_master(ac, sbr, &sbr->spectrum_params); 1033 if (err >= 0) 1034 err = sbr_make_f_derived(ac, sbr); 1035 if (err < 0) { 1036 av_log(ac->avctx, AV_LOG_ERROR, 1037 "SBR reset failed. Switching SBR to pure upsampling mode.\n"); 1038 sbr->start = 0; 1039 } 1040} 1041 1042/** 1043 * Decode Spectral Band Replication extension data; reference: table 4.55. 1044 * 1045 * @param crc flag indicating the presence of CRC checksum 1046 * @param cnt length of TYPE_FIL syntactic element in bytes 1047 * 1048 * @return Returns number of bytes consumed from the TYPE_FIL element. 1049 */ 1050int ff_decode_sbr_extension(AACContext *ac, SpectralBandReplication *sbr, 1051 GetBitContext *gb_host, int crc, int cnt, int id_aac) 1052{ 1053 unsigned int num_sbr_bits = 0, num_align_bits; 1054 unsigned bytes_read; 1055 GetBitContext gbc = *gb_host, *gb = &gbc; 1056 skip_bits_long(gb_host, cnt*8 - 4); 1057 1058 sbr->reset = 0; 1059 1060 if (!sbr->sample_rate) 1061 sbr->sample_rate = 2 * ac->m4ac.sample_rate; //TODO use the nominal sample rate for arbitrary sample rate support 1062 if (!ac->m4ac.ext_sample_rate) 1063 ac->m4ac.ext_sample_rate = 2 * ac->m4ac.sample_rate; 1064 1065 if (crc) { 1066 skip_bits(gb, 10); // bs_sbr_crc_bits; TODO - implement CRC check 1067 num_sbr_bits += 10; 1068 } 1069 1070 //Save some state from the previous frame. 1071 sbr->kx[0] = sbr->kx[1]; 1072 sbr->m[0] = sbr->m[1]; 1073 1074 num_sbr_bits++; 1075 if (get_bits1(gb)) // bs_header_flag 1076 num_sbr_bits += read_sbr_header(sbr, gb); 1077 1078 if (sbr->reset) 1079 sbr_reset(ac, sbr); 1080 1081 if (sbr->start) 1082 num_sbr_bits += read_sbr_data(ac, sbr, gb, id_aac); 1083 1084 num_align_bits = ((cnt << 3) - 4 - num_sbr_bits) & 7; 1085 bytes_read = ((num_sbr_bits + num_align_bits + 4) >> 3); 1086 1087 if (bytes_read > cnt) { 1088 av_log(ac->avctx, AV_LOG_ERROR, 1089 "Expected to read %d SBR bytes actually read %d.\n", cnt, bytes_read); 1090 } 1091 return cnt; 1092} 1093 1094/// Dequantization and stereo decoding (14496-3 sp04 p203) 1095static void sbr_dequant(SpectralBandReplication *sbr, int id_aac) 1096{ 1097 int k, e; 1098 int ch; 1099 1100 if (id_aac == TYPE_CPE && sbr->bs_coupling) { 1101 float alpha = sbr->data[0].bs_amp_res ? 1.0f : 0.5f; 1102 float pan_offset = sbr->data[0].bs_amp_res ? 12.0f : 24.0f; 1103 for (e = 1; e <= sbr->data[0].bs_num_env; e++) { 1104 for (k = 0; k < sbr->n[sbr->data[0].bs_freq_res[e]]; k++) { 1105 float temp1 = exp2f(sbr->data[0].env_facs[e][k] * alpha + 7.0f); 1106 float temp2 = exp2f((pan_offset - sbr->data[1].env_facs[e][k]) * alpha); 1107 float fac = temp1 / (1.0f + temp2); 1108 sbr->data[0].env_facs[e][k] = fac; 1109 sbr->data[1].env_facs[e][k] = fac * temp2; 1110 } 1111 } 1112 for (e = 1; e <= sbr->data[0].bs_num_noise; e++) { 1113 for (k = 0; k < sbr->n_q; k++) { 1114 float temp1 = exp2f(NOISE_FLOOR_OFFSET - sbr->data[0].noise_facs[e][k] + 1); 1115 float temp2 = exp2f(12 - sbr->data[1].noise_facs[e][k]); 1116 float fac = temp1 / (1.0f + temp2); 1117 sbr->data[0].noise_facs[e][k] = fac; 1118 sbr->data[1].noise_facs[e][k] = fac * temp2; 1119 } 1120 } 1121 } else { // SCE or one non-coupled CPE 1122 for (ch = 0; ch < (id_aac == TYPE_CPE) + 1; ch++) { 1123 float alpha = sbr->data[ch].bs_amp_res ? 1.0f : 0.5f; 1124 for (e = 1; e <= sbr->data[ch].bs_num_env; e++) 1125 for (k = 0; k < sbr->n[sbr->data[ch].bs_freq_res[e]]; k++) 1126 sbr->data[ch].env_facs[e][k] = 1127 exp2f(alpha * sbr->data[ch].env_facs[e][k] + 6.0f); 1128 for (e = 1; e <= sbr->data[ch].bs_num_noise; e++) 1129 for (k = 0; k < sbr->n_q; k++) 1130 sbr->data[ch].noise_facs[e][k] = 1131 exp2f(NOISE_FLOOR_OFFSET - sbr->data[ch].noise_facs[e][k]); 1132 } 1133 } 1134} 1135 1136/** 1137 * Analysis QMF Bank (14496-3 sp04 p206) 1138 * 1139 * @param x pointer to the beginning of the first sample window 1140 * @param W array of complex-valued samples split into subbands 1141 */ 1142static void sbr_qmf_analysis(DSPContext *dsp, FFTContext *mdct, const float *in, float *x, 1143 float z[320], float W[2][32][32][2]) 1144{ 1145 int i, k; 1146 memcpy(W[0], W[1], sizeof(W[0])); 1147 memcpy(x , x+1024, (320-32)*sizeof(x[0])); 1148 memcpy(x+288, in, 1024*sizeof(x[0])); 1149 for (i = 0; i < 32; i++) { // numTimeSlots*RATE = 16*2 as 960 sample frames 1150 // are not supported 1151 dsp->vector_fmul_reverse(z, sbr_qmf_window_ds, x, 320); 1152 for (k = 0; k < 64; k++) { 1153 float f = z[k] + z[k + 64] + z[k + 128] + z[k + 192] + z[k + 256]; 1154 z[k] = f; 1155 } 1156 //Shuffle to IMDCT 1157 z[64] = z[0]; 1158 for (k = 1; k < 32; k++) { 1159 z[64+2*k-1] = z[ k]; 1160 z[64+2*k ] = -z[64-k]; 1161 } 1162 z[64+63] = z[32]; 1163 1164 mdct->imdct_half(mdct, z, z+64); 1165 for (k = 0; k < 32; k++) { 1166 W[1][i][k][0] = -z[63-k]; 1167 W[1][i][k][1] = z[k]; 1168 } 1169 x += 32; 1170 } 1171} 1172 1173/** 1174 * Synthesis QMF Bank (14496-3 sp04 p206) and Downsampled Synthesis QMF Bank 1175 * (14496-3 sp04 p206) 1176 */ 1177static void sbr_qmf_synthesis(DSPContext *dsp, FFTContext *mdct, 1178 float *out, float X[2][38][64], 1179 float mdct_buf[2][64], 1180 float *v0, int *v_off, const unsigned int div) 1181{ 1182 int i, n; 1183 const float *sbr_qmf_window = div ? sbr_qmf_window_ds : sbr_qmf_window_us; 1184 const int step = 128 >> div; 1185 float *v; 1186 for (i = 0; i < 32; i++) { 1187 if (*v_off < step) { 1188 int saved_samples = (1280 - 128) >> div; 1189 memcpy(&v0[SBR_SYNTHESIS_BUF_SIZE - saved_samples], v0, saved_samples * sizeof(float)); 1190 *v_off = SBR_SYNTHESIS_BUF_SIZE - saved_samples - step; 1191 } else { 1192 *v_off -= step; 1193 } 1194 v = v0 + *v_off; 1195 if (div) { 1196 for (n = 0; n < 32; n++) { 1197 X[0][i][ n] = -X[0][i][n]; 1198 X[0][i][32+n] = X[1][i][31-n]; 1199 } 1200 mdct->imdct_half(mdct, mdct_buf[0], X[0][i]); 1201 for (n = 0; n < 32; n++) { 1202 v[ n] = mdct_buf[0][63 - 2*n]; 1203 v[63 - n] = -mdct_buf[0][62 - 2*n]; 1204 } 1205 } else { 1206 for (n = 1; n < 64; n+=2) { 1207 X[1][i][n] = -X[1][i][n]; 1208 } 1209 mdct->imdct_half(mdct, mdct_buf[0], X[0][i]); 1210 mdct->imdct_half(mdct, mdct_buf[1], X[1][i]); 1211 for (n = 0; n < 64; n++) { 1212 v[ n] = -mdct_buf[0][63 - n] + mdct_buf[1][ n ]; 1213 v[127 - n] = mdct_buf[0][63 - n] + mdct_buf[1][ n ]; 1214 } 1215 } 1216 dsp->vector_fmul_add(out, v , sbr_qmf_window , zero64, 64 >> div); 1217 dsp->vector_fmul_add(out, v + ( 192 >> div), sbr_qmf_window + ( 64 >> div), out , 64 >> div); 1218 dsp->vector_fmul_add(out, v + ( 256 >> div), sbr_qmf_window + (128 >> div), out , 64 >> div); 1219 dsp->vector_fmul_add(out, v + ( 448 >> div), sbr_qmf_window + (192 >> div), out , 64 >> div); 1220 dsp->vector_fmul_add(out, v + ( 512 >> div), sbr_qmf_window + (256 >> div), out , 64 >> div); 1221 dsp->vector_fmul_add(out, v + ( 704 >> div), sbr_qmf_window + (320 >> div), out , 64 >> div); 1222 dsp->vector_fmul_add(out, v + ( 768 >> div), sbr_qmf_window + (384 >> div), out , 64 >> div); 1223 dsp->vector_fmul_add(out, v + ( 960 >> div), sbr_qmf_window + (448 >> div), out , 64 >> div); 1224 dsp->vector_fmul_add(out, v + (1024 >> div), sbr_qmf_window + (512 >> div), out , 64 >> div); 1225 dsp->vector_fmul_add(out, v + (1216 >> div), sbr_qmf_window + (576 >> div), out , 64 >> div); 1226 out += 64 >> div; 1227 } 1228} 1229 1230static void autocorrelate(const float x[40][2], float phi[3][2][2], int lag) 1231{ 1232 int i; 1233 float real_sum = 0.0f; 1234 float imag_sum = 0.0f; 1235 if (lag) { 1236 for (i = 1; i < 38; i++) { 1237 real_sum += x[i][0] * x[i+lag][0] + x[i][1] * x[i+lag][1]; 1238 imag_sum += x[i][0] * x[i+lag][1] - x[i][1] * x[i+lag][0]; 1239 } 1240 phi[2-lag][1][0] = real_sum + x[ 0][0] * x[lag][0] + x[ 0][1] * x[lag][1]; 1241 phi[2-lag][1][1] = imag_sum + x[ 0][0] * x[lag][1] - x[ 0][1] * x[lag][0]; 1242 if (lag == 1) { 1243 phi[0][0][0] = real_sum + x[38][0] * x[39][0] + x[38][1] * x[39][1]; 1244 phi[0][0][1] = imag_sum + x[38][0] * x[39][1] - x[38][1] * x[39][0]; 1245 } 1246 } else { 1247 for (i = 1; i < 38; i++) { 1248 real_sum += x[i][0] * x[i][0] + x[i][1] * x[i][1]; 1249 } 1250 phi[2][1][0] = real_sum + x[ 0][0] * x[ 0][0] + x[ 0][1] * x[ 0][1]; 1251 phi[1][0][0] = real_sum + x[38][0] * x[38][0] + x[38][1] * x[38][1]; 1252 } 1253} 1254 1255/** High Frequency Generation (14496-3 sp04 p214+) and Inverse Filtering 1256 * (14496-3 sp04 p214) 1257 * Warning: This routine does not seem numerically stable. 1258 */ 1259static void sbr_hf_inverse_filter(float (*alpha0)[2], float (*alpha1)[2], 1260 const float X_low[32][40][2], int k0) 1261{ 1262 int k; 1263 for (k = 0; k < k0; k++) { 1264 float phi[3][2][2], dk; 1265 1266 autocorrelate(X_low[k], phi, 0); 1267 autocorrelate(X_low[k], phi, 1); 1268 autocorrelate(X_low[k], phi, 2); 1269 1270 dk = phi[2][1][0] * phi[1][0][0] - 1271 (phi[1][1][0] * phi[1][1][0] + phi[1][1][1] * phi[1][1][1]) / 1.000001f; 1272 1273 if (!dk) { 1274 alpha1[k][0] = 0; 1275 alpha1[k][1] = 0; 1276 } else { 1277 float temp_real, temp_im; 1278 temp_real = phi[0][0][0] * phi[1][1][0] - 1279 phi[0][0][1] * phi[1][1][1] - 1280 phi[0][1][0] * phi[1][0][0]; 1281 temp_im = phi[0][0][0] * phi[1][1][1] + 1282 phi[0][0][1] * phi[1][1][0] - 1283 phi[0][1][1] * phi[1][0][0]; 1284 1285 alpha1[k][0] = temp_real / dk; 1286 alpha1[k][1] = temp_im / dk; 1287 } 1288 1289 if (!phi[1][0][0]) { 1290 alpha0[k][0] = 0; 1291 alpha0[k][1] = 0; 1292 } else { 1293 float temp_real, temp_im; 1294 temp_real = phi[0][0][0] + alpha1[k][0] * phi[1][1][0] + 1295 alpha1[k][1] * phi[1][1][1]; 1296 temp_im = phi[0][0][1] + alpha1[k][1] * phi[1][1][0] - 1297 alpha1[k][0] * phi[1][1][1]; 1298 1299 alpha0[k][0] = -temp_real / phi[1][0][0]; 1300 alpha0[k][1] = -temp_im / phi[1][0][0]; 1301 } 1302 1303 if (alpha1[k][0] * alpha1[k][0] + alpha1[k][1] * alpha1[k][1] >= 16.0f || 1304 alpha0[k][0] * alpha0[k][0] + alpha0[k][1] * alpha0[k][1] >= 16.0f) { 1305 alpha1[k][0] = 0; 1306 alpha1[k][1] = 0; 1307 alpha0[k][0] = 0; 1308 alpha0[k][1] = 0; 1309 } 1310 } 1311} 1312 1313/// Chirp Factors (14496-3 sp04 p214) 1314static void sbr_chirp(SpectralBandReplication *sbr, SBRData *ch_data) 1315{ 1316 int i; 1317 float new_bw; 1318 static const float bw_tab[] = { 0.0f, 0.75f, 0.9f, 0.98f }; 1319 1320 for (i = 0; i < sbr->n_q; i++) { 1321 if (ch_data->bs_invf_mode[0][i] + ch_data->bs_invf_mode[1][i] == 1) { 1322 new_bw = 0.6f; 1323 } else 1324 new_bw = bw_tab[ch_data->bs_invf_mode[0][i]]; 1325 1326 if (new_bw < ch_data->bw_array[i]) { 1327 new_bw = 0.75f * new_bw + 0.25f * ch_data->bw_array[i]; 1328 } else 1329 new_bw = 0.90625f * new_bw + 0.09375f * ch_data->bw_array[i]; 1330 ch_data->bw_array[i] = new_bw < 0.015625f ? 0.0f : new_bw; 1331 } 1332} 1333 1334/// Generate the subband filtered lowband 1335static int sbr_lf_gen(AACContext *ac, SpectralBandReplication *sbr, 1336 float X_low[32][40][2], const float W[2][32][32][2]) 1337{ 1338 int i, k; 1339 const int t_HFGen = 8; 1340 const int i_f = 32; 1341 memset(X_low, 0, 32*sizeof(*X_low)); 1342 for (k = 0; k < sbr->kx[1]; k++) { 1343 for (i = t_HFGen; i < i_f + t_HFGen; i++) { 1344 X_low[k][i][0] = W[1][i - t_HFGen][k][0]; 1345 X_low[k][i][1] = W[1][i - t_HFGen][k][1]; 1346 } 1347 } 1348 for (k = 0; k < sbr->kx[0]; k++) { 1349 for (i = 0; i < t_HFGen; i++) { 1350 X_low[k][i][0] = W[0][i + i_f - t_HFGen][k][0]; 1351 X_low[k][i][1] = W[0][i + i_f - t_HFGen][k][1]; 1352 } 1353 } 1354 return 0; 1355} 1356 1357/// High Frequency Generator (14496-3 sp04 p215) 1358static int sbr_hf_gen(AACContext *ac, SpectralBandReplication *sbr, 1359 float X_high[64][40][2], const float X_low[32][40][2], 1360 const float (*alpha0)[2], const float (*alpha1)[2], 1361 const float bw_array[5], const uint8_t *t_env, 1362 int bs_num_env) 1363{ 1364 int i, j, x; 1365 int g = 0; 1366 int k = sbr->kx[1]; 1367 for (j = 0; j < sbr->num_patches; j++) { 1368 for (x = 0; x < sbr->patch_num_subbands[j]; x++, k++) { 1369 float alpha[4]; 1370 const int p = sbr->patch_start_subband[j] + x; 1371 while (g <= sbr->n_q && k >= sbr->f_tablenoise[g]) 1372 g++; 1373 g--; 1374 1375 if (g < 0) { 1376 av_log(ac->avctx, AV_LOG_ERROR, 1377 "ERROR : no subband found for frequency %d\n", k); 1378 return -1; 1379 } 1380 1381 alpha[0] = alpha1[p][0] * bw_array[g] * bw_array[g]; 1382 alpha[1] = alpha1[p][1] * bw_array[g] * bw_array[g]; 1383 alpha[2] = alpha0[p][0] * bw_array[g]; 1384 alpha[3] = alpha0[p][1] * bw_array[g]; 1385 1386 for (i = 2 * t_env[0]; i < 2 * t_env[bs_num_env]; i++) { 1387 const int idx = i + ENVELOPE_ADJUSTMENT_OFFSET; 1388 X_high[k][idx][0] = 1389 X_low[p][idx - 2][0] * alpha[0] - 1390 X_low[p][idx - 2][1] * alpha[1] + 1391 X_low[p][idx - 1][0] * alpha[2] - 1392 X_low[p][idx - 1][1] * alpha[3] + 1393 X_low[p][idx][0]; 1394 X_high[k][idx][1] = 1395 X_low[p][idx - 2][1] * alpha[0] + 1396 X_low[p][idx - 2][0] * alpha[1] + 1397 X_low[p][idx - 1][1] * alpha[2] + 1398 X_low[p][idx - 1][0] * alpha[3] + 1399 X_low[p][idx][1]; 1400 } 1401 } 1402 } 1403 if (k < sbr->m[1] + sbr->kx[1]) 1404 memset(X_high + k, 0, (sbr->m[1] + sbr->kx[1] - k) * sizeof(*X_high)); 1405 1406 return 0; 1407} 1408 1409/// Generate the subband filtered lowband 1410static int sbr_x_gen(SpectralBandReplication *sbr, float X[2][38][64], 1411 const float X_low[32][40][2], const float Y[2][38][64][2], 1412 int ch) 1413{ 1414 int k, i; 1415 const int i_f = 32; 1416 const int i_Temp = FFMAX(2*sbr->data[ch].t_env_num_env_old - i_f, 0); 1417 memset(X, 0, 2*sizeof(*X)); 1418 for (k = 0; k < sbr->kx[0]; k++) { 1419 for (i = 0; i < i_Temp; i++) { 1420 X[0][i][k] = X_low[k][i + ENVELOPE_ADJUSTMENT_OFFSET][0]; 1421 X[1][i][k] = X_low[k][i + ENVELOPE_ADJUSTMENT_OFFSET][1]; 1422 } 1423 } 1424 for (; k < sbr->kx[0] + sbr->m[0]; k++) { 1425 for (i = 0; i < i_Temp; i++) { 1426 X[0][i][k] = Y[0][i + i_f][k][0]; 1427 X[1][i][k] = Y[0][i + i_f][k][1]; 1428 } 1429 } 1430 1431 for (k = 0; k < sbr->kx[1]; k++) { 1432 for (i = i_Temp; i < 38; i++) { 1433 X[0][i][k] = X_low[k][i + ENVELOPE_ADJUSTMENT_OFFSET][0]; 1434 X[1][i][k] = X_low[k][i + ENVELOPE_ADJUSTMENT_OFFSET][1]; 1435 } 1436 } 1437 for (; k < sbr->kx[1] + sbr->m[1]; k++) { 1438 for (i = i_Temp; i < i_f; i++) { 1439 X[0][i][k] = Y[1][i][k][0]; 1440 X[1][i][k] = Y[1][i][k][1]; 1441 } 1442 } 1443 return 0; 1444} 1445 1446/** High Frequency Adjustment (14496-3 sp04 p217) and Mapping 1447 * (14496-3 sp04 p217) 1448 */ 1449static void sbr_mapping(AACContext *ac, SpectralBandReplication *sbr, 1450 SBRData *ch_data, int e_a[2]) 1451{ 1452 int e, i, m; 1453 1454 memset(ch_data->s_indexmapped[1], 0, 7*sizeof(ch_data->s_indexmapped[1])); 1455 for (e = 0; e < ch_data->bs_num_env; e++) { 1456 const unsigned int ilim = sbr->n[ch_data->bs_freq_res[e + 1]]; 1457 uint16_t *table = ch_data->bs_freq_res[e + 1] ? sbr->f_tablehigh : sbr->f_tablelow; 1458 int k; 1459 1460 for (i = 0; i < ilim; i++) 1461 for (m = table[i]; m < table[i + 1]; m++) 1462 sbr->e_origmapped[e][m - sbr->kx[1]] = ch_data->env_facs[e+1][i]; 1463 1464 // ch_data->bs_num_noise > 1 => 2 noise floors 1465 k = (ch_data->bs_num_noise > 1) && (ch_data->t_env[e] >= ch_data->t_q[1]); 1466 for (i = 0; i < sbr->n_q; i++) 1467 for (m = sbr->f_tablenoise[i]; m < sbr->f_tablenoise[i + 1]; m++) 1468 sbr->q_mapped[e][m - sbr->kx[1]] = ch_data->noise_facs[k+1][i]; 1469 1470 for (i = 0; i < sbr->n[1]; i++) { 1471 if (ch_data->bs_add_harmonic_flag) { 1472 const unsigned int m_midpoint = 1473 (sbr->f_tablehigh[i] + sbr->f_tablehigh[i + 1]) >> 1; 1474 1475 ch_data->s_indexmapped[e + 1][m_midpoint - sbr->kx[1]] = ch_data->bs_add_harmonic[i] * 1476 (e >= e_a[1] || (ch_data->s_indexmapped[0][m_midpoint - sbr->kx[1]] == 1)); 1477 } 1478 } 1479 1480 for (i = 0; i < ilim; i++) { 1481 int additional_sinusoid_present = 0; 1482 for (m = table[i]; m < table[i + 1]; m++) { 1483 if (ch_data->s_indexmapped[e + 1][m - sbr->kx[1]]) { 1484 additional_sinusoid_present = 1; 1485 break; 1486 } 1487 } 1488 memset(&sbr->s_mapped[e][table[i] - sbr->kx[1]], additional_sinusoid_present, 1489 (table[i + 1] - table[i]) * sizeof(sbr->s_mapped[e][0])); 1490 } 1491 } 1492 1493 memcpy(ch_data->s_indexmapped[0], ch_data->s_indexmapped[ch_data->bs_num_env], sizeof(ch_data->s_indexmapped[0])); 1494} 1495 1496/// Estimation of current envelope (14496-3 sp04 p218) 1497static void sbr_env_estimate(float (*e_curr)[48], float X_high[64][40][2], 1498 SpectralBandReplication *sbr, SBRData *ch_data) 1499{ 1500 int e, i, m; 1501 1502 if (sbr->bs_interpol_freq) { 1503 for (e = 0; e < ch_data->bs_num_env; e++) { 1504 const float recip_env_size = 0.5f / (ch_data->t_env[e + 1] - ch_data->t_env[e]); 1505 int ilb = ch_data->t_env[e] * 2 + ENVELOPE_ADJUSTMENT_OFFSET; 1506 int iub = ch_data->t_env[e + 1] * 2 + ENVELOPE_ADJUSTMENT_OFFSET; 1507 1508 for (m = 0; m < sbr->m[1]; m++) { 1509 float sum = 0.0f; 1510 1511 for (i = ilb; i < iub; i++) { 1512 sum += X_high[m + sbr->kx[1]][i][0] * X_high[m + sbr->kx[1]][i][0] + 1513 X_high[m + sbr->kx[1]][i][1] * X_high[m + sbr->kx[1]][i][1]; 1514 } 1515 e_curr[e][m] = sum * recip_env_size; 1516 } 1517 } 1518 } else { 1519 int k, p; 1520 1521 for (e = 0; e < ch_data->bs_num_env; e++) { 1522 const int env_size = 2 * (ch_data->t_env[e + 1] - ch_data->t_env[e]); 1523 int ilb = ch_data->t_env[e] * 2 + ENVELOPE_ADJUSTMENT_OFFSET; 1524 int iub = ch_data->t_env[e + 1] * 2 + ENVELOPE_ADJUSTMENT_OFFSET; 1525 const uint16_t *table = ch_data->bs_freq_res[e + 1] ? sbr->f_tablehigh : sbr->f_tablelow; 1526 1527 for (p = 0; p < sbr->n[ch_data->bs_freq_res[e + 1]]; p++) { 1528 float sum = 0.0f; 1529 const int den = env_size * (table[p + 1] - table[p]); 1530 1531 for (k = table[p]; k < table[p + 1]; k++) { 1532 for (i = ilb; i < iub; i++) { 1533 sum += X_high[k][i][0] * X_high[k][i][0] + 1534 X_high[k][i][1] * X_high[k][i][1]; 1535 } 1536 } 1537 sum /= den; 1538 for (k = table[p]; k < table[p + 1]; k++) { 1539 e_curr[e][k - sbr->kx[1]] = sum; 1540 } 1541 } 1542 } 1543 } 1544} 1545 1546/** 1547 * Calculation of levels of additional HF signal components (14496-3 sp04 p219) 1548 * and Calculation of gain (14496-3 sp04 p219) 1549 */ 1550static void sbr_gain_calc(AACContext *ac, SpectralBandReplication *sbr, 1551 SBRData *ch_data, const int e_a[2]) 1552{ 1553 int e, k, m; 1554 // max gain limits : -3dB, 0dB, 3dB, inf dB (limiter off) 1555 static const float limgain[4] = { 0.70795, 1.0, 1.41254, 10000000000 }; 1556 1557 for (e = 0; e < ch_data->bs_num_env; e++) { 1558 int delta = !((e == e_a[1]) || (e == e_a[0])); 1559 for (k = 0; k < sbr->n_lim; k++) { 1560 float gain_boost, gain_max; 1561 float sum[2] = { 0.0f, 0.0f }; 1562 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) { 1563 const float temp = sbr->e_origmapped[e][m] / (1.0f + sbr->q_mapped[e][m]); 1564 sbr->q_m[e][m] = sqrtf(temp * sbr->q_mapped[e][m]); 1565 sbr->s_m[e][m] = sqrtf(temp * ch_data->s_indexmapped[e + 1][m]); 1566 if (!sbr->s_mapped[e][m]) { 1567 sbr->gain[e][m] = sqrtf(sbr->e_origmapped[e][m] / 1568 ((1.0f + sbr->e_curr[e][m]) * 1569 (1.0f + sbr->q_mapped[e][m] * delta))); 1570 } else { 1571 sbr->gain[e][m] = sqrtf(sbr->e_origmapped[e][m] * sbr->q_mapped[e][m] / 1572 ((1.0f + sbr->e_curr[e][m]) * 1573 (1.0f + sbr->q_mapped[e][m]))); 1574 } 1575 } 1576 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) { 1577 sum[0] += sbr->e_origmapped[e][m]; 1578 sum[1] += sbr->e_curr[e][m]; 1579 } 1580 gain_max = limgain[sbr->bs_limiter_gains] * sqrtf((FLT_EPSILON + sum[0]) / (FLT_EPSILON + sum[1])); 1581 gain_max = FFMIN(100000.f, gain_max); 1582 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) { 1583 float q_m_max = sbr->q_m[e][m] * gain_max / sbr->gain[e][m]; 1584 sbr->q_m[e][m] = FFMIN(sbr->q_m[e][m], q_m_max); 1585 sbr->gain[e][m] = FFMIN(sbr->gain[e][m], gain_max); 1586 } 1587 sum[0] = sum[1] = 0.0f; 1588 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) { 1589 sum[0] += sbr->e_origmapped[e][m]; 1590 sum[1] += sbr->e_curr[e][m] * sbr->gain[e][m] * sbr->gain[e][m] 1591 + sbr->s_m[e][m] * sbr->s_m[e][m] 1592 + (delta && !sbr->s_m[e][m]) * sbr->q_m[e][m] * sbr->q_m[e][m]; 1593 } 1594 gain_boost = sqrtf((FLT_EPSILON + sum[0]) / (FLT_EPSILON + sum[1])); 1595 gain_boost = FFMIN(1.584893192f, gain_boost); 1596 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) { 1597 sbr->gain[e][m] *= gain_boost; 1598 sbr->q_m[e][m] *= gain_boost; 1599 sbr->s_m[e][m] *= gain_boost; 1600 } 1601 } 1602 } 1603} 1604 1605/// Assembling HF Signals (14496-3 sp04 p220) 1606static void sbr_hf_assemble(float Y[2][38][64][2], const float X_high[64][40][2], 1607 SpectralBandReplication *sbr, SBRData *ch_data, 1608 const int e_a[2]) 1609{ 1610 int e, i, j, m; 1611 const int h_SL = 4 * !sbr->bs_smoothing_mode; 1612 const int kx = sbr->kx[1]; 1613 const int m_max = sbr->m[1]; 1614 static const float h_smooth[5] = { 1615 0.33333333333333, 1616 0.30150283239582, 1617 0.21816949906249, 1618 0.11516383427084, 1619 0.03183050093751, 1620 }; 1621 static const int8_t phi[2][4] = { 1622 { 1, 0, -1, 0}, // real 1623 { 0, 1, 0, -1}, // imaginary 1624 }; 1625 float (*g_temp)[48] = ch_data->g_temp, (*q_temp)[48] = ch_data->q_temp; 1626 int indexnoise = ch_data->f_indexnoise; 1627 int indexsine = ch_data->f_indexsine; 1628 memcpy(Y[0], Y[1], sizeof(Y[0])); 1629 1630 if (sbr->reset) { 1631 for (i = 0; i < h_SL; i++) { 1632 memcpy(g_temp[i + 2*ch_data->t_env[0]], sbr->gain[0], m_max * sizeof(sbr->gain[0][0])); 1633 memcpy(q_temp[i + 2*ch_data->t_env[0]], sbr->q_m[0], m_max * sizeof(sbr->q_m[0][0])); 1634 } 1635 } else if (h_SL) { 1636 memcpy(g_temp[2*ch_data->t_env[0]], g_temp[2*ch_data->t_env_num_env_old], 4*sizeof(g_temp[0])); 1637 memcpy(q_temp[2*ch_data->t_env[0]], q_temp[2*ch_data->t_env_num_env_old], 4*sizeof(q_temp[0])); 1638 } 1639 1640 for (e = 0; e < ch_data->bs_num_env; e++) { 1641 for (i = 2 * ch_data->t_env[e]; i < 2 * ch_data->t_env[e + 1]; i++) { 1642 memcpy(g_temp[h_SL + i], sbr->gain[e], m_max * sizeof(sbr->gain[0][0])); 1643 memcpy(q_temp[h_SL + i], sbr->q_m[e], m_max * sizeof(sbr->q_m[0][0])); 1644 } 1645 } 1646 1647 for (e = 0; e < ch_data->bs_num_env; e++) { 1648 for (i = 2 * ch_data->t_env[e]; i < 2 * ch_data->t_env[e + 1]; i++) { 1649 int phi_sign = (1 - 2*(kx & 1)); 1650 1651 if (h_SL && e != e_a[0] && e != e_a[1]) { 1652 for (m = 0; m < m_max; m++) { 1653 const int idx1 = i + h_SL; 1654 float g_filt = 0.0f; 1655 for (j = 0; j <= h_SL; j++) 1656 g_filt += g_temp[idx1 - j][m] * h_smooth[j]; 1657 Y[1][i][m + kx][0] = 1658 X_high[m + kx][i + ENVELOPE_ADJUSTMENT_OFFSET][0] * g_filt; 1659 Y[1][i][m + kx][1] = 1660 X_high[m + kx][i + ENVELOPE_ADJUSTMENT_OFFSET][1] * g_filt; 1661 } 1662 } else { 1663 for (m = 0; m < m_max; m++) { 1664 const float g_filt = g_temp[i + h_SL][m]; 1665 Y[1][i][m + kx][0] = 1666 X_high[m + kx][i + ENVELOPE_ADJUSTMENT_OFFSET][0] * g_filt; 1667 Y[1][i][m + kx][1] = 1668 X_high[m + kx][i + ENVELOPE_ADJUSTMENT_OFFSET][1] * g_filt; 1669 } 1670 } 1671 1672 if (e != e_a[0] && e != e_a[1]) { 1673 for (m = 0; m < m_max; m++) { 1674 indexnoise = (indexnoise + 1) & 0x1ff; 1675 if (sbr->s_m[e][m]) { 1676 Y[1][i][m + kx][0] += 1677 sbr->s_m[e][m] * phi[0][indexsine]; 1678 Y[1][i][m + kx][1] += 1679 sbr->s_m[e][m] * (phi[1][indexsine] * phi_sign); 1680 } else { 1681 float q_filt; 1682 if (h_SL) { 1683 const int idx1 = i + h_SL; 1684 q_filt = 0.0f; 1685 for (j = 0; j <= h_SL; j++) 1686 q_filt += q_temp[idx1 - j][m] * h_smooth[j]; 1687 } else { 1688 q_filt = q_temp[i][m]; 1689 } 1690 Y[1][i][m + kx][0] += 1691 q_filt * sbr_noise_table[indexnoise][0]; 1692 Y[1][i][m + kx][1] += 1693 q_filt * sbr_noise_table[indexnoise][1]; 1694 } 1695 phi_sign = -phi_sign; 1696 } 1697 } else { 1698 indexnoise = (indexnoise + m_max) & 0x1ff; 1699 for (m = 0; m < m_max; m++) { 1700 Y[1][i][m + kx][0] += 1701 sbr->s_m[e][m] * phi[0][indexsine]; 1702 Y[1][i][m + kx][1] += 1703 sbr->s_m[e][m] * (phi[1][indexsine] * phi_sign); 1704 phi_sign = -phi_sign; 1705 } 1706 } 1707 indexsine = (indexsine + 1) & 3; 1708 } 1709 } 1710 ch_data->f_indexnoise = indexnoise; 1711 ch_data->f_indexsine = indexsine; 1712} 1713 1714void ff_sbr_apply(AACContext *ac, SpectralBandReplication *sbr, int id_aac, 1715 float* L, float* R) 1716{ 1717 int downsampled = ac->m4ac.ext_sample_rate < sbr->sample_rate; 1718 int ch; 1719 int nch = (id_aac == TYPE_CPE) ? 2 : 1; 1720 1721 if (sbr->start) { 1722 sbr_dequant(sbr, id_aac); 1723 } 1724 for (ch = 0; ch < nch; ch++) { 1725 /* decode channel */ 1726 sbr_qmf_analysis(&ac->dsp, &sbr->mdct_ana, ch ? R : L, sbr->data[ch].analysis_filterbank_samples, 1727 (float*)sbr->qmf_filter_scratch, 1728 sbr->data[ch].W); 1729 sbr_lf_gen(ac, sbr, sbr->X_low, sbr->data[ch].W); 1730 if (sbr->start) { 1731 sbr_hf_inverse_filter(sbr->alpha0, sbr->alpha1, sbr->X_low, sbr->k[0]); 1732 sbr_chirp(sbr, &sbr->data[ch]); 1733 sbr_hf_gen(ac, sbr, sbr->X_high, sbr->X_low, sbr->alpha0, sbr->alpha1, 1734 sbr->data[ch].bw_array, sbr->data[ch].t_env, 1735 sbr->data[ch].bs_num_env); 1736 1737 // hf_adj 1738 sbr_mapping(ac, sbr, &sbr->data[ch], sbr->data[ch].e_a); 1739 sbr_env_estimate(sbr->e_curr, sbr->X_high, sbr, &sbr->data[ch]); 1740 sbr_gain_calc(ac, sbr, &sbr->data[ch], sbr->data[ch].e_a); 1741 sbr_hf_assemble(sbr->data[ch].Y, sbr->X_high, sbr, &sbr->data[ch], 1742 sbr->data[ch].e_a); 1743 } 1744 1745 /* synthesis */ 1746 sbr_x_gen(sbr, sbr->X[ch], sbr->X_low, sbr->data[ch].Y, ch); 1747 } 1748 1749 if (ac->m4ac.ps == 1) { 1750 if (sbr->ps.start) { 1751 ff_ps_apply(ac->avctx, &sbr->ps, sbr->X[0], sbr->X[1], sbr->kx[1] + sbr->m[1]); 1752 } else { 1753 memcpy(sbr->X[1], sbr->X[0], sizeof(sbr->X[0])); 1754 } 1755 nch = 2; 1756 } 1757 1758 sbr_qmf_synthesis(&ac->dsp, &sbr->mdct, L, sbr->X[0], sbr->qmf_filter_scratch, 1759 sbr->data[0].synthesis_filterbank_samples, 1760 &sbr->data[0].synthesis_filterbank_samples_offset, 1761 downsampled); 1762 if (nch == 2) 1763 sbr_qmf_synthesis(&ac->dsp, &sbr->mdct, R, sbr->X[1], sbr->qmf_filter_scratch, 1764 sbr->data[1].synthesis_filterbank_samples, 1765 &sbr->data[1].synthesis_filterbank_samples_offset, 1766 downsampled); 1767} 1768