You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

422 lines
15KB

  1. /*
  2. * TwinVQ decoder
  3. * Copyright (c) 2009 Vitor Sessak
  4. *
  5. * This file is part of Libav.
  6. *
  7. * Libav 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. * Libav 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 Libav; if not, write to the Free Software
  19. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20. */
  21. #include <math.h>
  22. #include <stdint.h>
  23. #include "libavutil/channel_layout.h"
  24. #include "avcodec.h"
  25. #include "get_bits.h"
  26. #include "internal.h"
  27. #include "twinvq.h"
  28. #include "twinvq_data.h"
  29. static const TwinVQModeTab mode_08_08 = {
  30. {
  31. { 8, bark_tab_s08_64, 10, tab.fcb08s, 1, 5, tab.cb0808s0, tab.cb0808s1, 18 },
  32. { 2, bark_tab_m08_256, 20, tab.fcb08m, 2, 5, tab.cb0808m0, tab.cb0808m1, 16 },
  33. { 1, bark_tab_l08_512, 30, tab.fcb08l, 3, 6, tab.cb0808l0, tab.cb0808l1, 17 }
  34. },
  35. 512, 12, tab.lsp08, 1, 5, 3, 3, tab.shape08, 8, 28, 20, 6, 40
  36. };
  37. static const TwinVQModeTab mode_11_08 = {
  38. {
  39. { 8, bark_tab_s11_64, 10, tab.fcb11s, 1, 5, tab.cb1108s0, tab.cb1108s1, 29 },
  40. { 2, bark_tab_m11_256, 20, tab.fcb11m, 2, 5, tab.cb1108m0, tab.cb1108m1, 24 },
  41. { 1, bark_tab_l11_512, 30, tab.fcb11l, 3, 6, tab.cb1108l0, tab.cb1108l1, 27 }
  42. },
  43. 512, 16, tab.lsp11, 1, 6, 4, 3, tab.shape11, 9, 36, 30, 7, 90
  44. };
  45. static const TwinVQModeTab mode_11_10 = {
  46. {
  47. { 8, bark_tab_s11_64, 10, tab.fcb11s, 1, 5, tab.cb1110s0, tab.cb1110s1, 21 },
  48. { 2, bark_tab_m11_256, 20, tab.fcb11m, 2, 5, tab.cb1110m0, tab.cb1110m1, 18 },
  49. { 1, bark_tab_l11_512, 30, tab.fcb11l, 3, 6, tab.cb1110l0, tab.cb1110l1, 20 }
  50. },
  51. 512, 16, tab.lsp11, 1, 6, 4, 3, tab.shape11, 9, 36, 30, 7, 90
  52. };
  53. static const TwinVQModeTab mode_16_16 = {
  54. {
  55. { 8, bark_tab_s16_128, 10, tab.fcb16s, 1, 5, tab.cb1616s0, tab.cb1616s1, 16 },
  56. { 2, bark_tab_m16_512, 20, tab.fcb16m, 2, 5, tab.cb1616m0, tab.cb1616m1, 15 },
  57. { 1, bark_tab_l16_1024, 30, tab.fcb16l, 3, 6, tab.cb1616l0, tab.cb1616l1, 16 }
  58. },
  59. 1024, 16, tab.lsp16, 1, 6, 4, 3, tab.shape16, 9, 56, 60, 7, 180
  60. };
  61. static const TwinVQModeTab mode_22_20 = {
  62. {
  63. { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2220s0, tab.cb2220s1, 18 },
  64. { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2220m0, tab.cb2220m1, 17 },
  65. { 1, bark_tab_l22_1024, 32, tab.fcb22l_1, 4, 6, tab.cb2220l0, tab.cb2220l1, 18 }
  66. },
  67. 1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
  68. };
  69. static const TwinVQModeTab mode_22_24 = {
  70. {
  71. { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2224s0, tab.cb2224s1, 15 },
  72. { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2224m0, tab.cb2224m1, 14 },
  73. { 1, bark_tab_l22_1024, 32, tab.fcb22l_1, 4, 6, tab.cb2224l0, tab.cb2224l1, 15 }
  74. },
  75. 1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
  76. };
  77. static const TwinVQModeTab mode_22_32 = {
  78. {
  79. { 4, bark_tab_s22_128, 10, tab.fcb22s_2, 1, 6, tab.cb2232s0, tab.cb2232s1, 11 },
  80. { 2, bark_tab_m22_256, 20, tab.fcb22m_2, 2, 6, tab.cb2232m0, tab.cb2232m1, 11 },
  81. { 1, bark_tab_l22_512, 32, tab.fcb22l_2, 4, 6, tab.cb2232l0, tab.cb2232l1, 12 }
  82. },
  83. 512, 16, tab.lsp22_2, 1, 6, 4, 4, tab.shape22_2, 9, 56, 36, 7, 72
  84. };
  85. static const TwinVQModeTab mode_44_40 = {
  86. {
  87. { 16, bark_tab_s44_128, 10, tab.fcb44s, 1, 6, tab.cb4440s0, tab.cb4440s1, 18 },
  88. { 4, bark_tab_m44_512, 20, tab.fcb44m, 2, 6, tab.cb4440m0, tab.cb4440m1, 17 },
  89. { 1, bark_tab_l44_2048, 40, tab.fcb44l, 4, 6, tab.cb4440l0, tab.cb4440l1, 17 }
  90. },
  91. 2048, 20, tab.lsp44, 1, 6, 4, 4, tab.shape44, 9, 84, 54, 7, 432
  92. };
  93. static const TwinVQModeTab mode_44_48 = {
  94. {
  95. { 16, bark_tab_s44_128, 10, tab.fcb44s, 1, 6, tab.cb4448s0, tab.cb4448s1, 15 },
  96. { 4, bark_tab_m44_512, 20, tab.fcb44m, 2, 6, tab.cb4448m0, tab.cb4448m1, 14 },
  97. { 1, bark_tab_l44_2048, 40, tab.fcb44l, 4, 6, tab.cb4448l0, tab.cb4448l1, 14 }
  98. },
  99. 2048, 20, tab.lsp44, 1, 6, 4, 4, tab.shape44, 9, 84, 54, 7, 432
  100. };
  101. /**
  102. * Evaluate a * b / 400 rounded to the nearest integer. When, for example,
  103. * a * b == 200 and the nearest integer is ill-defined, use a table to emulate
  104. * the following broken float-based implementation used by the binary decoder:
  105. *
  106. * @code
  107. * static int very_broken_op(int a, int b)
  108. * {
  109. * static float test; // Ugh, force gcc to do the division first...
  110. *
  111. * test = a / 400.0;
  112. * return b * test + 0.5;
  113. * }
  114. * @endcode
  115. *
  116. * @note if this function is replaced by just ROUNDED_DIV(a * b, 400.0), the
  117. * stddev between the original file (before encoding with Yamaha encoder) and
  118. * the decoded output increases, which leads one to believe that the encoder
  119. * expects exactly this broken calculation.
  120. */
  121. static int very_broken_op(int a, int b)
  122. {
  123. int x = a * b + 200;
  124. int size;
  125. const uint8_t *rtab;
  126. if (x % 400 || b % 5)
  127. return x / 400;
  128. x /= 400;
  129. size = tabs[b / 5].size;
  130. rtab = tabs[b / 5].tab;
  131. return x - rtab[size * av_log2(2 * (x - 1) / size) + (x - 1) % size];
  132. }
  133. /**
  134. * Sum to data a periodic peak of a given period, width and shape.
  135. *
  136. * @param period the period of the peak divised by 400.0
  137. */
  138. static void add_peak(int period, int width, const float *shape,
  139. float ppc_gain, float *speech, int len)
  140. {
  141. int i, j;
  142. const float *shape_end = shape + len;
  143. int center;
  144. // First peak centered around zero
  145. for (i = 0; i < width / 2; i++)
  146. speech[i] += ppc_gain * *shape++;
  147. for (i = 1; i < ROUNDED_DIV(len, width); i++) {
  148. center = very_broken_op(period, i);
  149. for (j = -width / 2; j < (width + 1) / 2; j++)
  150. speech[j + center] += ppc_gain * *shape++;
  151. }
  152. // For the last block, be careful not to go beyond the end of the buffer
  153. center = very_broken_op(period, i);
  154. for (j = -width / 2; j < (width + 1) / 2 && shape < shape_end; j++)
  155. speech[j + center] += ppc_gain * *shape++;
  156. }
  157. static void decode_ppc(TwinVQContext *tctx, int period_coef, int g_coef,
  158. const float *shape, float *speech)
  159. {
  160. const TwinVQModeTab *mtab = tctx->mtab;
  161. int isampf = tctx->avctx->sample_rate / 1000;
  162. int ibps = tctx->avctx->bit_rate / (1000 * tctx->avctx->channels);
  163. int min_period = ROUNDED_DIV(40 * 2 * mtab->size, isampf);
  164. int max_period = ROUNDED_DIV(40 * 2 * mtab->size * 6, isampf);
  165. int period_range = max_period - min_period;
  166. float pgain_step = 25000.0 / ((1 << mtab->pgain_bit) - 1);
  167. float ppc_gain = 1.0 / 8192 *
  168. twinvq_mulawinv(pgain_step * g_coef +
  169. pgain_step / 2,
  170. 25000.0, TWINVQ_PGAIN_MU);
  171. // This is actually the period multiplied by 400. It is just linearly coded
  172. // between its maximum and minimum value.
  173. int period = min_period +
  174. ROUNDED_DIV(period_coef * period_range,
  175. (1 << mtab->ppc_period_bit) - 1);
  176. int width;
  177. if (isampf == 22 && ibps == 32) {
  178. // For some unknown reason, NTT decided to code this case differently...
  179. width = ROUNDED_DIV((period + 800) * mtab->peak_per2wid,
  180. 400 * mtab->size);
  181. } else
  182. width = period * mtab->peak_per2wid / (400 * mtab->size);
  183. add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len);
  184. }
  185. static void dec_bark_env(TwinVQContext *tctx, const uint8_t *in, int use_hist,
  186. int ch, float *out, float gain,
  187. enum TwinVQFrameType ftype)
  188. {
  189. const TwinVQModeTab *mtab = tctx->mtab;
  190. int i, j;
  191. float *hist = tctx->bark_hist[ftype][ch];
  192. float val = ((const float []) { 0.4, 0.35, 0.28 })[ftype];
  193. int bark_n_coef = mtab->fmode[ftype].bark_n_coef;
  194. int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef;
  195. int idx = 0;
  196. for (i = 0; i < fw_cb_len; i++)
  197. for (j = 0; j < bark_n_coef; j++, idx++) {
  198. float tmp2 = mtab->fmode[ftype].bark_cb[fw_cb_len * in[j] + i] *
  199. (1.0 / 4096);
  200. float st = use_hist ? (1.0 - val) * tmp2 + val * hist[idx] + 1.0
  201. : tmp2 + 1.0;
  202. hist[idx] = tmp2;
  203. if (st < -1.0)
  204. st = 1.0;
  205. twinvq_memset_float(out, st * gain, mtab->fmode[ftype].bark_tab[idx]);
  206. out += mtab->fmode[ftype].bark_tab[idx];
  207. }
  208. }
  209. static void read_cb_data(TwinVQContext *tctx, GetBitContext *gb,
  210. uint8_t *dst, enum TwinVQFrameType ftype)
  211. {
  212. int i;
  213. for (i = 0; i < tctx->n_div[ftype]; i++) {
  214. int bs_second_part = (i >= tctx->bits_main_spec_change[ftype]);
  215. *dst++ = get_bits(gb, tctx->bits_main_spec[0][ftype][bs_second_part]);
  216. *dst++ = get_bits(gb, tctx->bits_main_spec[1][ftype][bs_second_part]);
  217. }
  218. }
  219. static int twinvq_read_bitstream(AVCodecContext *avctx, TwinVQContext *tctx,
  220. const uint8_t *buf, int buf_size)
  221. {
  222. TwinVQFrameData *bits = &tctx->bits;
  223. const TwinVQModeTab *mtab = tctx->mtab;
  224. int channels = tctx->avctx->channels;
  225. int sub;
  226. GetBitContext gb;
  227. int i, j, k;
  228. if (buf_size * 8 < avctx->bit_rate * mtab->size / avctx->sample_rate + 8) {
  229. av_log(avctx, AV_LOG_ERROR,
  230. "Frame too small (%d bytes). Truncated file?\n", buf_size);
  231. return AVERROR(EINVAL);
  232. }
  233. init_get_bits(&gb, buf, buf_size * 8);
  234. skip_bits(&gb, get_bits(&gb, 8));
  235. bits->window_type = get_bits(&gb, TWINVQ_WINDOW_TYPE_BITS);
  236. if (bits->window_type > 8) {
  237. av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
  238. return AVERROR_INVALIDDATA;
  239. }
  240. bits->ftype = ff_twinvq_wtype_to_ftype_table[tctx->bits.window_type];
  241. sub = mtab->fmode[bits->ftype].sub;
  242. read_cb_data(tctx, &gb, bits->main_coeffs, bits->ftype);
  243. for (i = 0; i < channels; i++)
  244. for (j = 0; j < sub; j++)
  245. for (k = 0; k < mtab->fmode[bits->ftype].bark_n_coef; k++)
  246. bits->bark1[i][j][k] =
  247. get_bits(&gb, mtab->fmode[bits->ftype].bark_n_bit);
  248. for (i = 0; i < channels; i++)
  249. for (j = 0; j < sub; j++)
  250. bits->bark_use_hist[i][j] = get_bits1(&gb);
  251. if (bits->ftype == TWINVQ_FT_LONG) {
  252. for (i = 0; i < channels; i++)
  253. bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS);
  254. } else {
  255. for (i = 0; i < channels; i++) {
  256. bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS);
  257. for (j = 0; j < sub; j++)
  258. bits->sub_gain_bits[i * sub + j] = get_bits(&gb,
  259. TWINVQ_SUB_GAIN_BITS);
  260. }
  261. }
  262. for (i = 0; i < channels; i++) {
  263. bits->lpc_hist_idx[i] = get_bits(&gb, mtab->lsp_bit0);
  264. bits->lpc_idx1[i] = get_bits(&gb, mtab->lsp_bit1);
  265. for (j = 0; j < mtab->lsp_split; j++)
  266. bits->lpc_idx2[i][j] = get_bits(&gb, mtab->lsp_bit2);
  267. }
  268. if (bits->ftype == TWINVQ_FT_LONG) {
  269. read_cb_data(tctx, &gb, bits->ppc_coeffs, 3);
  270. for (i = 0; i < channels; i++) {
  271. bits->p_coef[i] = get_bits(&gb, mtab->ppc_period_bit);
  272. bits->g_coef[i] = get_bits(&gb, mtab->pgain_bit);
  273. }
  274. }
  275. return 0;
  276. }
  277. static av_cold int twinvq_decode_init(AVCodecContext *avctx)
  278. {
  279. int isampf, ibps;
  280. TwinVQContext *tctx = avctx->priv_data;
  281. if (!avctx->extradata || avctx->extradata_size < 12) {
  282. av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n");
  283. return AVERROR_INVALIDDATA;
  284. }
  285. avctx->channels = AV_RB32(avctx->extradata) + 1;
  286. avctx->bit_rate = AV_RB32(avctx->extradata + 4) * 1000;
  287. isampf = AV_RB32(avctx->extradata + 8);
  288. if (isampf < 8 || isampf > 44) {
  289. av_log(avctx, AV_LOG_ERROR, "Unsupported sample rate\n");
  290. return AVERROR_INVALIDDATA;
  291. }
  292. switch (isampf) {
  293. case 44:
  294. avctx->sample_rate = 44100;
  295. break;
  296. case 22:
  297. avctx->sample_rate = 22050;
  298. break;
  299. case 11:
  300. avctx->sample_rate = 11025;
  301. break;
  302. default:
  303. avctx->sample_rate = isampf * 1000;
  304. break;
  305. }
  306. if (avctx->channels <= 0 || avctx->channels > TWINVQ_CHANNELS_MAX) {
  307. av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n",
  308. avctx->channels);
  309. return -1;
  310. }
  311. avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO
  312. : AV_CH_LAYOUT_STEREO;
  313. ibps = avctx->bit_rate / (1000 * avctx->channels);
  314. switch ((isampf << 8) + ibps) {
  315. case (8 << 8) + 8:
  316. tctx->mtab = &mode_08_08;
  317. break;
  318. case (11 << 8) + 8:
  319. tctx->mtab = &mode_11_08;
  320. break;
  321. case (11 << 8) + 10:
  322. tctx->mtab = &mode_11_10;
  323. break;
  324. case (16 << 8) + 16:
  325. tctx->mtab = &mode_16_16;
  326. break;
  327. case (22 << 8) + 20:
  328. tctx->mtab = &mode_22_20;
  329. break;
  330. case (22 << 8) + 24:
  331. tctx->mtab = &mode_22_24;
  332. break;
  333. case (22 << 8) + 32:
  334. tctx->mtab = &mode_22_32;
  335. break;
  336. case (44 << 8) + 40:
  337. tctx->mtab = &mode_44_40;
  338. break;
  339. case (44 << 8) + 48:
  340. tctx->mtab = &mode_44_48;
  341. break;
  342. default:
  343. av_log(avctx, AV_LOG_ERROR,
  344. "This version does not support %d kHz - %d kbit/s/ch mode.\n",
  345. isampf, isampf);
  346. return -1;
  347. }
  348. tctx->codec = TWINVQ_CODEC_VQF;
  349. tctx->read_bitstream = twinvq_read_bitstream;
  350. tctx->dec_bark_env = dec_bark_env;
  351. tctx->decode_ppc = decode_ppc;
  352. return ff_twinvq_decode_init(avctx);
  353. }
  354. AVCodec ff_twinvq_decoder = {
  355. .name = "twinvq",
  356. .type = AVMEDIA_TYPE_AUDIO,
  357. .id = AV_CODEC_ID_TWINVQ,
  358. .priv_data_size = sizeof(TwinVQContext),
  359. .init = twinvq_decode_init,
  360. .close = ff_twinvq_decode_close,
  361. .decode = ff_twinvq_decode_frame,
  362. .capabilities = CODEC_CAP_DR1,
  363. .long_name = NULL_IF_CONFIG_SMALL("VQF TwinVQ"),
  364. .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
  365. AV_SAMPLE_FMT_NONE },
  366. };