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.

823 lines
25KB

  1. /*
  2. * QCELP decoder
  3. * Copyright (c) 2007 Reynaldo H. Verdejo Pinochet
  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. * @file libavcodec/qcelpdec.c
  23. * QCELP decoder
  24. * @author Reynaldo H. Verdejo Pinochet
  25. * @remark FFmpeg merging spearheaded by Kenan Gillet
  26. * @remark Development mentored by Benjamin Larson
  27. */
  28. #include <stddef.h>
  29. #include "avcodec.h"
  30. #include "internal.h"
  31. #include "get_bits.h"
  32. #include "qcelpdata.h"
  33. #include "celp_math.h"
  34. #include "celp_filters.h"
  35. #include "acelp_vectors.h"
  36. #include "lsp.h"
  37. #undef NDEBUG
  38. #include <assert.h>
  39. typedef enum
  40. {
  41. I_F_Q = -1, /*!< insufficient frame quality */
  42. SILENCE,
  43. RATE_OCTAVE,
  44. RATE_QUARTER,
  45. RATE_HALF,
  46. RATE_FULL
  47. } qcelp_packet_rate;
  48. typedef struct
  49. {
  50. GetBitContext gb;
  51. qcelp_packet_rate bitrate;
  52. QCELPFrame frame; /*!< unpacked data frame */
  53. uint8_t erasure_count;
  54. uint8_t octave_count; /*!< count the consecutive RATE_OCTAVE frames */
  55. float prev_lspf[10];
  56. float predictor_lspf[10];/*!< LSP predictor for RATE_OCTAVE and I_F_Q */
  57. float pitch_synthesis_filter_mem[303];
  58. float pitch_pre_filter_mem[303];
  59. float rnd_fir_filter_mem[180];
  60. float formant_mem[170];
  61. float last_codebook_gain;
  62. int prev_g1[2];
  63. int prev_bitrate;
  64. float pitch_gain[4];
  65. uint8_t pitch_lag[4];
  66. uint16_t first16bits;
  67. uint8_t warned_buf_mismatch_bitrate;
  68. } QCELPContext;
  69. /**
  70. * Initialize the speech codec according to the specification.
  71. *
  72. * TIA/EIA/IS-733 2.4.9
  73. */
  74. static av_cold int qcelp_decode_init(AVCodecContext *avctx)
  75. {
  76. QCELPContext *q = avctx->priv_data;
  77. int i;
  78. avctx->sample_fmt = SAMPLE_FMT_FLT;
  79. for(i=0; i<10; i++)
  80. q->prev_lspf[i] = (i+1)/11.;
  81. return 0;
  82. }
  83. /**
  84. * Decodes the 10 quantized LSP frequencies from the LSPV/LSP
  85. * transmission codes of any bitrate and checks for badly received packets.
  86. *
  87. * @param q the context
  88. * @param lspf line spectral pair frequencies
  89. *
  90. * @return 0 on success, -1 if the packet is badly received
  91. *
  92. * TIA/EIA/IS-733 2.4.3.2.6.2-2, 2.4.8.7.3
  93. */
  94. static int decode_lspf(QCELPContext *q, float *lspf)
  95. {
  96. int i;
  97. float tmp_lspf, smooth, erasure_coeff;
  98. const float *predictors;
  99. if(q->bitrate == RATE_OCTAVE || q->bitrate == I_F_Q)
  100. {
  101. predictors = (q->prev_bitrate != RATE_OCTAVE &&
  102. q->prev_bitrate != I_F_Q ?
  103. q->prev_lspf : q->predictor_lspf);
  104. if(q->bitrate == RATE_OCTAVE)
  105. {
  106. q->octave_count++;
  107. for(i=0; i<10; i++)
  108. {
  109. q->predictor_lspf[i] =
  110. lspf[i] = (q->frame.lspv[i] ? QCELP_LSP_SPREAD_FACTOR
  111. : -QCELP_LSP_SPREAD_FACTOR)
  112. + predictors[i] * QCELP_LSP_OCTAVE_PREDICTOR
  113. + (i + 1) * ((1 - QCELP_LSP_OCTAVE_PREDICTOR)/11);
  114. }
  115. smooth = (q->octave_count < 10 ? .875 : 0.1);
  116. }else
  117. {
  118. erasure_coeff = QCELP_LSP_OCTAVE_PREDICTOR;
  119. assert(q->bitrate == I_F_Q);
  120. if(q->erasure_count > 1)
  121. erasure_coeff *= (q->erasure_count < 4 ? 0.9 : 0.7);
  122. for(i=0; i<10; i++)
  123. {
  124. q->predictor_lspf[i] =
  125. lspf[i] = (i + 1) * ( 1 - erasure_coeff)/11
  126. + erasure_coeff * predictors[i];
  127. }
  128. smooth = 0.125;
  129. }
  130. // Check the stability of the LSP frequencies.
  131. lspf[0] = FFMAX(lspf[0], QCELP_LSP_SPREAD_FACTOR);
  132. for(i=1; i<10; i++)
  133. lspf[i] = FFMAX(lspf[i], (lspf[i-1] + QCELP_LSP_SPREAD_FACTOR));
  134. lspf[9] = FFMIN(lspf[9], (1.0 - QCELP_LSP_SPREAD_FACTOR));
  135. for(i=9; i>0; i--)
  136. lspf[i-1] = FFMIN(lspf[i-1], (lspf[i] - QCELP_LSP_SPREAD_FACTOR));
  137. // Low-pass filter the LSP frequencies.
  138. ff_weighted_vector_sumf(lspf, lspf, q->prev_lspf, smooth, 1.0-smooth, 10);
  139. }else
  140. {
  141. q->octave_count = 0;
  142. tmp_lspf = 0.;
  143. for(i=0; i<5 ; i++)
  144. {
  145. lspf[2*i+0] = tmp_lspf += qcelp_lspvq[i][q->frame.lspv[i]][0] * 0.0001;
  146. lspf[2*i+1] = tmp_lspf += qcelp_lspvq[i][q->frame.lspv[i]][1] * 0.0001;
  147. }
  148. // Check for badly received packets.
  149. if(q->bitrate == RATE_QUARTER)
  150. {
  151. if(lspf[9] <= .70 || lspf[9] >= .97)
  152. return -1;
  153. for(i=3; i<10; i++)
  154. if(fabs(lspf[i] - lspf[i-2]) < .08)
  155. return -1;
  156. }else
  157. {
  158. if(lspf[9] <= .66 || lspf[9] >= .985)
  159. return -1;
  160. for(i=4; i<10; i++)
  161. if (fabs(lspf[i] - lspf[i-4]) < .0931)
  162. return -1;
  163. }
  164. }
  165. return 0;
  166. }
  167. /**
  168. * Converts codebook transmission codes to GAIN and INDEX.
  169. *
  170. * @param q the context
  171. * @param gain array holding the decoded gain
  172. *
  173. * TIA/EIA/IS-733 2.4.6.2
  174. */
  175. static void decode_gain_and_index(QCELPContext *q,
  176. float *gain) {
  177. int i, subframes_count, g1[16];
  178. float slope;
  179. if(q->bitrate >= RATE_QUARTER)
  180. {
  181. switch(q->bitrate)
  182. {
  183. case RATE_FULL: subframes_count = 16; break;
  184. case RATE_HALF: subframes_count = 4; break;
  185. default: subframes_count = 5;
  186. }
  187. for(i=0; i<subframes_count; i++)
  188. {
  189. g1[i] = 4 * q->frame.cbgain[i];
  190. if(q->bitrate == RATE_FULL && !((i+1) & 3))
  191. {
  192. g1[i] += av_clip((g1[i-1] + g1[i-2] + g1[i-3]) / 3 - 6, 0, 32);
  193. }
  194. gain[i] = qcelp_g12ga[g1[i]];
  195. if(q->frame.cbsign[i])
  196. {
  197. gain[i] = -gain[i];
  198. q->frame.cindex[i] = (q->frame.cindex[i]-89) & 127;
  199. }
  200. }
  201. q->prev_g1[0] = g1[i-2];
  202. q->prev_g1[1] = g1[i-1];
  203. q->last_codebook_gain = qcelp_g12ga[g1[i-1]];
  204. if(q->bitrate == RATE_QUARTER)
  205. {
  206. // Provide smoothing of the unvoiced excitation energy.
  207. gain[7] = gain[4];
  208. gain[6] = 0.4*gain[3] + 0.6*gain[4];
  209. gain[5] = gain[3];
  210. gain[4] = 0.8*gain[2] + 0.2*gain[3];
  211. gain[3] = 0.2*gain[1] + 0.8*gain[2];
  212. gain[2] = gain[1];
  213. gain[1] = 0.6*gain[0] + 0.4*gain[1];
  214. }
  215. }else if (q->bitrate != SILENCE)
  216. {
  217. if(q->bitrate == RATE_OCTAVE)
  218. {
  219. g1[0] = 2 * q->frame.cbgain[0]
  220. + av_clip((q->prev_g1[0] + q->prev_g1[1]) / 2 - 5, 0, 54);
  221. subframes_count = 8;
  222. }else
  223. {
  224. assert(q->bitrate == I_F_Q);
  225. g1[0] = q->prev_g1[1];
  226. switch(q->erasure_count)
  227. {
  228. case 1 : break;
  229. case 2 : g1[0] -= 1; break;
  230. case 3 : g1[0] -= 2; break;
  231. default: g1[0] -= 6;
  232. }
  233. if(g1[0] < 0)
  234. g1[0] = 0;
  235. subframes_count = 4;
  236. }
  237. // This interpolation is done to produce smoother background noise.
  238. slope = 0.5*(qcelp_g12ga[g1[0]] - q->last_codebook_gain) / subframes_count;
  239. for(i=1; i<=subframes_count; i++)
  240. gain[i-1] = q->last_codebook_gain + slope * i;
  241. q->last_codebook_gain = gain[i-2];
  242. q->prev_g1[0] = q->prev_g1[1];
  243. q->prev_g1[1] = g1[0];
  244. }
  245. }
  246. /**
  247. * If the received packet is Rate 1/4 a further sanity check is made of the
  248. * codebook gain.
  249. *
  250. * @param cbgain the unpacked cbgain array
  251. * @return -1 if the sanity check fails, 0 otherwise
  252. *
  253. * TIA/EIA/IS-733 2.4.8.7.3
  254. */
  255. static int codebook_sanity_check_for_rate_quarter(const uint8_t *cbgain)
  256. {
  257. int i, diff, prev_diff=0;
  258. for(i=1; i<5; i++)
  259. {
  260. diff = cbgain[i] - cbgain[i-1];
  261. if(FFABS(diff) > 10)
  262. return -1;
  263. else if(FFABS(diff - prev_diff) > 12)
  264. return -1;
  265. prev_diff = diff;
  266. }
  267. return 0;
  268. }
  269. /**
  270. * Computes the scaled codebook vector Cdn From INDEX and GAIN
  271. * for all rates.
  272. *
  273. * The specification lacks some information here.
  274. *
  275. * TIA/EIA/IS-733 has an omission on the codebook index determination
  276. * formula for RATE_FULL and RATE_HALF frames at section 2.4.8.1.1. It says
  277. * you have to subtract the decoded index parameter from the given scaled
  278. * codebook vector index 'n' to get the desired circular codebook index, but
  279. * it does not mention that you have to clamp 'n' to [0-9] in order to get
  280. * RI-compliant results.
  281. *
  282. * The reason for this mistake seems to be the fact they forgot to mention you
  283. * have to do these calculations per codebook subframe and adjust given
  284. * equation values accordingly.
  285. *
  286. * @param q the context
  287. * @param gain array holding the 4 pitch subframe gain values
  288. * @param cdn_vector array for the generated scaled codebook vector
  289. */
  290. static void compute_svector(QCELPContext *q, const float *gain,
  291. float *cdn_vector)
  292. {
  293. int i, j, k;
  294. uint16_t cbseed, cindex;
  295. float *rnd, tmp_gain, fir_filter_value;
  296. switch(q->bitrate)
  297. {
  298. case RATE_FULL:
  299. for(i=0; i<16; i++)
  300. {
  301. tmp_gain = gain[i] * QCELP_RATE_FULL_CODEBOOK_RATIO;
  302. cindex = -q->frame.cindex[i];
  303. for(j=0; j<10; j++)
  304. *cdn_vector++ = tmp_gain * qcelp_rate_full_codebook[cindex++ & 127];
  305. }
  306. break;
  307. case RATE_HALF:
  308. for(i=0; i<4; i++)
  309. {
  310. tmp_gain = gain[i] * QCELP_RATE_HALF_CODEBOOK_RATIO;
  311. cindex = -q->frame.cindex[i];
  312. for (j = 0; j < 40; j++)
  313. *cdn_vector++ = tmp_gain * qcelp_rate_half_codebook[cindex++ & 127];
  314. }
  315. break;
  316. case RATE_QUARTER:
  317. cbseed = (0x0003 & q->frame.lspv[4])<<14 |
  318. (0x003F & q->frame.lspv[3])<< 8 |
  319. (0x0060 & q->frame.lspv[2])<< 1 |
  320. (0x0007 & q->frame.lspv[1])<< 3 |
  321. (0x0038 & q->frame.lspv[0])>> 3 ;
  322. rnd = q->rnd_fir_filter_mem + 20;
  323. for(i=0; i<8; i++)
  324. {
  325. tmp_gain = gain[i] * (QCELP_SQRT1887 / 32768.0);
  326. for(k=0; k<20; k++)
  327. {
  328. cbseed = 521 * cbseed + 259;
  329. *rnd = (int16_t)cbseed;
  330. // FIR filter
  331. fir_filter_value = 0.0;
  332. for(j=0; j<10; j++)
  333. fir_filter_value += qcelp_rnd_fir_coefs[j ]
  334. * (rnd[-j ] + rnd[-20+j]);
  335. fir_filter_value += qcelp_rnd_fir_coefs[10] * rnd[-10];
  336. *cdn_vector++ = tmp_gain * fir_filter_value;
  337. rnd++;
  338. }
  339. }
  340. memcpy(q->rnd_fir_filter_mem, q->rnd_fir_filter_mem + 160, 20 * sizeof(float));
  341. break;
  342. case RATE_OCTAVE:
  343. cbseed = q->first16bits;
  344. for(i=0; i<8; i++)
  345. {
  346. tmp_gain = gain[i] * (QCELP_SQRT1887 / 32768.0);
  347. for(j=0; j<20; j++)
  348. {
  349. cbseed = 521 * cbseed + 259;
  350. *cdn_vector++ = tmp_gain * (int16_t)cbseed;
  351. }
  352. }
  353. break;
  354. case I_F_Q:
  355. cbseed = -44; // random codebook index
  356. for(i=0; i<4; i++)
  357. {
  358. tmp_gain = gain[i] * QCELP_RATE_FULL_CODEBOOK_RATIO;
  359. for(j=0; j<40; j++)
  360. *cdn_vector++ = tmp_gain * qcelp_rate_full_codebook[cbseed++ & 127];
  361. }
  362. break;
  363. case SILENCE:
  364. memset(cdn_vector, 0, 160 * sizeof(float));
  365. break;
  366. }
  367. }
  368. /**
  369. * Apply generic gain control.
  370. *
  371. * @param v_out output vector
  372. * @param v_in gain-controlled vector
  373. * @param v_ref vector to control gain of
  374. *
  375. * TIA/EIA/IS-733 2.4.8.3, 2.4.8.6
  376. */
  377. static void apply_gain_ctrl(float *v_out, const float *v_ref,
  378. const float *v_in)
  379. {
  380. int i;
  381. for (i = 0; i < 160; i += 40)
  382. ff_scale_vector_to_given_sum_of_squares(v_out + i, v_in + i,
  383. ff_dot_productf(v_ref + i,
  384. v_ref + i, 40),
  385. 40);
  386. }
  387. /**
  388. * Apply filter in pitch-subframe steps.
  389. *
  390. * @param memory buffer for the previous state of the filter
  391. * - must be able to contain 303 elements
  392. * - the 143 first elements are from the previous state
  393. * - the next 160 are for output
  394. * @param v_in input filter vector
  395. * @param gain per-subframe gain array, each element is between 0.0 and 2.0
  396. * @param lag per-subframe lag array, each element is
  397. * - between 16 and 143 if its corresponding pfrac is 0,
  398. * - between 16 and 139 otherwise
  399. * @param pfrac per-subframe boolean array, 1 if the lag is fractional, 0
  400. * otherwise
  401. *
  402. * @return filter output vector
  403. */
  404. static const float *do_pitchfilter(float memory[303], const float v_in[160],
  405. const float gain[4], const uint8_t *lag,
  406. const uint8_t pfrac[4])
  407. {
  408. int i, j;
  409. float *v_lag, *v_out;
  410. const float *v_len;
  411. v_out = memory + 143; // Output vector starts at memory[143].
  412. for(i=0; i<4; i++)
  413. {
  414. if(gain[i])
  415. {
  416. v_lag = memory + 143 + 40 * i - lag[i];
  417. for(v_len=v_in+40; v_in<v_len; v_in++)
  418. {
  419. if(pfrac[i]) // If it is a fractional lag...
  420. {
  421. for(j=0, *v_out=0.; j<4; j++)
  422. *v_out += qcelp_hammsinc_table[j] * (v_lag[j-4] + v_lag[3-j]);
  423. }else
  424. *v_out = *v_lag;
  425. *v_out = *v_in + gain[i] * *v_out;
  426. v_lag++;
  427. v_out++;
  428. }
  429. }else
  430. {
  431. memcpy(v_out, v_in, 40 * sizeof(float));
  432. v_in += 40;
  433. v_out += 40;
  434. }
  435. }
  436. memmove(memory, memory + 160, 143 * sizeof(float));
  437. return memory + 143;
  438. }
  439. /**
  440. * Apply pitch synthesis filter and pitch prefilter to the scaled codebook vector.
  441. * TIA/EIA/IS-733 2.4.5.2, 2.4.8.7.2
  442. *
  443. * @param q the context
  444. * @param cdn_vector the scaled codebook vector
  445. */
  446. static void apply_pitch_filters(QCELPContext *q, float *cdn_vector)
  447. {
  448. int i;
  449. const float *v_synthesis_filtered, *v_pre_filtered;
  450. if(q->bitrate >= RATE_HALF ||
  451. q->bitrate == SILENCE ||
  452. (q->bitrate == I_F_Q && (q->prev_bitrate >= RATE_HALF)))
  453. {
  454. if(q->bitrate >= RATE_HALF)
  455. {
  456. // Compute gain & lag for the whole frame.
  457. for(i=0; i<4; i++)
  458. {
  459. q->pitch_gain[i] = q->frame.plag[i] ? (q->frame.pgain[i] + 1) * 0.25 : 0.0;
  460. q->pitch_lag[i] = q->frame.plag[i] + 16;
  461. }
  462. }else
  463. {
  464. float max_pitch_gain;
  465. if (q->bitrate == I_F_Q)
  466. {
  467. if (q->erasure_count < 3)
  468. max_pitch_gain = 0.9 - 0.3 * (q->erasure_count - 1);
  469. else
  470. max_pitch_gain = 0.0;
  471. }else
  472. {
  473. assert(q->bitrate == SILENCE);
  474. max_pitch_gain = 1.0;
  475. }
  476. for(i=0; i<4; i++)
  477. q->pitch_gain[i] = FFMIN(q->pitch_gain[i], max_pitch_gain);
  478. memset(q->frame.pfrac, 0, sizeof(q->frame.pfrac));
  479. }
  480. // pitch synthesis filter
  481. v_synthesis_filtered = do_pitchfilter(q->pitch_synthesis_filter_mem,
  482. cdn_vector, q->pitch_gain,
  483. q->pitch_lag, q->frame.pfrac);
  484. // pitch prefilter update
  485. for(i=0; i<4; i++)
  486. q->pitch_gain[i] = 0.5 * FFMIN(q->pitch_gain[i], 1.0);
  487. v_pre_filtered = do_pitchfilter(q->pitch_pre_filter_mem,
  488. v_synthesis_filtered,
  489. q->pitch_gain, q->pitch_lag,
  490. q->frame.pfrac);
  491. apply_gain_ctrl(cdn_vector, v_synthesis_filtered, v_pre_filtered);
  492. }else
  493. {
  494. memcpy(q->pitch_synthesis_filter_mem, cdn_vector + 17,
  495. 143 * sizeof(float));
  496. memcpy(q->pitch_pre_filter_mem, cdn_vector + 17, 143 * sizeof(float));
  497. memset(q->pitch_gain, 0, sizeof(q->pitch_gain));
  498. memset(q->pitch_lag, 0, sizeof(q->pitch_lag));
  499. }
  500. }
  501. /**
  502. * Reconstructs LPC coefficients from the line spectral pair frequencies
  503. * and performs bandwidth expansion.
  504. *
  505. * @param lspf line spectral pair frequencies
  506. * @param lpc linear predictive coding coefficients
  507. *
  508. * @note: bandwidth_expansion_coeff could be precalculated into a table
  509. * but it seems to be slower on x86
  510. *
  511. * TIA/EIA/IS-733 2.4.3.3.5
  512. */
  513. static void lspf2lpc(const float *lspf, float *lpc)
  514. {
  515. double lsp[10];
  516. double bandwidth_expansion_coeff = QCELP_BANDWIDTH_EXPANSION_COEFF;
  517. int i;
  518. for (i=0; i<10; i++)
  519. lsp[i] = cos(M_PI * lspf[i]);
  520. ff_acelp_lspd2lpc(lsp, lpc, 5);
  521. for (i=0; i<10; i++)
  522. {
  523. lpc[i] *= bandwidth_expansion_coeff;
  524. bandwidth_expansion_coeff *= QCELP_BANDWIDTH_EXPANSION_COEFF;
  525. }
  526. }
  527. /**
  528. * Interpolates LSP frequencies and computes LPC coefficients
  529. * for a given bitrate & pitch subframe.
  530. *
  531. * TIA/EIA/IS-733 2.4.3.3.4, 2.4.8.7.2
  532. *
  533. * @param q the context
  534. * @param curr_lspf LSP frequencies vector of the current frame
  535. * @param lpc float vector for the resulting LPC
  536. * @param subframe_num frame number in decoded stream
  537. */
  538. static void interpolate_lpc(QCELPContext *q, const float *curr_lspf,
  539. float *lpc, const int subframe_num)
  540. {
  541. float interpolated_lspf[10];
  542. float weight;
  543. if(q->bitrate >= RATE_QUARTER)
  544. weight = 0.25 * (subframe_num + 1);
  545. else if(q->bitrate == RATE_OCTAVE && !subframe_num)
  546. weight = 0.625;
  547. else
  548. weight = 1.0;
  549. if(weight != 1.0)
  550. {
  551. ff_weighted_vector_sumf(interpolated_lspf, curr_lspf, q->prev_lspf,
  552. weight, 1.0 - weight, 10);
  553. lspf2lpc(interpolated_lspf, lpc);
  554. }else if(q->bitrate >= RATE_QUARTER ||
  555. (q->bitrate == I_F_Q && !subframe_num))
  556. lspf2lpc(curr_lspf, lpc);
  557. else if(q->bitrate == SILENCE && !subframe_num)
  558. lspf2lpc(q->prev_lspf, lpc);
  559. }
  560. static qcelp_packet_rate buf_size2bitrate(const int buf_size)
  561. {
  562. switch(buf_size)
  563. {
  564. case 35: return RATE_FULL;
  565. case 17: return RATE_HALF;
  566. case 8: return RATE_QUARTER;
  567. case 4: return RATE_OCTAVE;
  568. case 1: return SILENCE;
  569. }
  570. return I_F_Q;
  571. }
  572. /**
  573. * Determine the bitrate from the frame size and/or the first byte of the frame.
  574. *
  575. * @param avctx the AV codec context
  576. * @param buf_size length of the buffer
  577. * @param buf the bufffer
  578. *
  579. * @return the bitrate on success,
  580. * I_F_Q if the bitrate cannot be satisfactorily determined
  581. *
  582. * TIA/EIA/IS-733 2.4.8.7.1
  583. */
  584. static qcelp_packet_rate determine_bitrate(AVCodecContext *avctx, const int buf_size,
  585. const uint8_t **buf)
  586. {
  587. qcelp_packet_rate bitrate;
  588. if((bitrate = buf_size2bitrate(buf_size)) >= 0)
  589. {
  590. if(bitrate > **buf)
  591. {
  592. QCELPContext *q = avctx->priv_data;
  593. if (!q->warned_buf_mismatch_bitrate)
  594. {
  595. av_log(avctx, AV_LOG_WARNING,
  596. "Claimed bitrate and buffer size mismatch.\n");
  597. q->warned_buf_mismatch_bitrate = 1;
  598. }
  599. bitrate = **buf;
  600. }else if(bitrate < **buf)
  601. {
  602. av_log(avctx, AV_LOG_ERROR,
  603. "Buffer is too small for the claimed bitrate.\n");
  604. return I_F_Q;
  605. }
  606. (*buf)++;
  607. }else if((bitrate = buf_size2bitrate(buf_size + 1)) >= 0)
  608. {
  609. av_log(avctx, AV_LOG_WARNING,
  610. "Bitrate byte is missing, guessing the bitrate from packet size.\n");
  611. }else
  612. return I_F_Q;
  613. if(bitrate == SILENCE)
  614. {
  615. //FIXME: Remove experimental warning when tested with samples.
  616. av_log_ask_for_sample(avctx, "'Blank frame handling is experimental.");
  617. }
  618. return bitrate;
  619. }
  620. static void warn_insufficient_frame_quality(AVCodecContext *avctx,
  621. const char *message)
  622. {
  623. av_log(avctx, AV_LOG_WARNING, "Frame #%d, IFQ: %s\n", avctx->frame_number,
  624. message);
  625. }
  626. static int qcelp_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
  627. AVPacket *avpkt)
  628. {
  629. const uint8_t *buf = avpkt->data;
  630. int buf_size = avpkt->size;
  631. QCELPContext *q = avctx->priv_data;
  632. float *outbuffer = data;
  633. int i;
  634. float quantized_lspf[10], lpc[10];
  635. float gain[16];
  636. float *formant_mem;
  637. if((q->bitrate = determine_bitrate(avctx, buf_size, &buf)) == I_F_Q)
  638. {
  639. warn_insufficient_frame_quality(avctx, "bitrate cannot be determined.");
  640. goto erasure;
  641. }
  642. if(q->bitrate == RATE_OCTAVE &&
  643. (q->first16bits = AV_RB16(buf)) == 0xFFFF)
  644. {
  645. warn_insufficient_frame_quality(avctx, "Bitrate is 1/8 and first 16 bits are on.");
  646. goto erasure;
  647. }
  648. if(q->bitrate > SILENCE)
  649. {
  650. const QCELPBitmap *bitmaps = qcelp_unpacking_bitmaps_per_rate[q->bitrate];
  651. const QCELPBitmap *bitmaps_end = qcelp_unpacking_bitmaps_per_rate[q->bitrate]
  652. + qcelp_unpacking_bitmaps_lengths[q->bitrate];
  653. uint8_t *unpacked_data = (uint8_t *)&q->frame;
  654. init_get_bits(&q->gb, buf, 8*buf_size);
  655. memset(&q->frame, 0, sizeof(QCELPFrame));
  656. for(; bitmaps < bitmaps_end; bitmaps++)
  657. unpacked_data[bitmaps->index] |= get_bits(&q->gb, bitmaps->bitlen) << bitmaps->bitpos;
  658. // Check for erasures/blanks on rates 1, 1/4 and 1/8.
  659. if(q->frame.reserved)
  660. {
  661. warn_insufficient_frame_quality(avctx, "Wrong data in reserved frame area.");
  662. goto erasure;
  663. }
  664. if(q->bitrate == RATE_QUARTER &&
  665. codebook_sanity_check_for_rate_quarter(q->frame.cbgain))
  666. {
  667. warn_insufficient_frame_quality(avctx, "Codebook gain sanity check failed.");
  668. goto erasure;
  669. }
  670. if(q->bitrate >= RATE_HALF)
  671. {
  672. for(i=0; i<4; i++)
  673. {
  674. if(q->frame.pfrac[i] && q->frame.plag[i] >= 124)
  675. {
  676. warn_insufficient_frame_quality(avctx, "Cannot initialize pitch filter.");
  677. goto erasure;
  678. }
  679. }
  680. }
  681. }
  682. decode_gain_and_index(q, gain);
  683. compute_svector(q, gain, outbuffer);
  684. if(decode_lspf(q, quantized_lspf) < 0)
  685. {
  686. warn_insufficient_frame_quality(avctx, "Badly received packets in frame.");
  687. goto erasure;
  688. }
  689. apply_pitch_filters(q, outbuffer);
  690. if(q->bitrate == I_F_Q)
  691. {
  692. erasure:
  693. q->bitrate = I_F_Q;
  694. q->erasure_count++;
  695. decode_gain_and_index(q, gain);
  696. compute_svector(q, gain, outbuffer);
  697. decode_lspf(q, quantized_lspf);
  698. apply_pitch_filters(q, outbuffer);
  699. }else
  700. q->erasure_count = 0;
  701. formant_mem = q->formant_mem + 10;
  702. for(i=0; i<4; i++)
  703. {
  704. interpolate_lpc(q, quantized_lspf, lpc, i);
  705. ff_celp_lp_synthesis_filterf(formant_mem, lpc, outbuffer + i * 40, 40,
  706. 10);
  707. formant_mem += 40;
  708. }
  709. memcpy(q->formant_mem, q->formant_mem + 160, 10 * sizeof(float));
  710. // FIXME: postfilter and final gain control should be here.
  711. // TIA/EIA/IS-733 2.4.8.6
  712. formant_mem = q->formant_mem + 10;
  713. for(i=0; i<160; i++)
  714. *outbuffer++ = av_clipf(*formant_mem++, QCELP_CLIP_LOWER_BOUND,
  715. QCELP_CLIP_UPPER_BOUND);
  716. memcpy(q->prev_lspf, quantized_lspf, sizeof(q->prev_lspf));
  717. q->prev_bitrate = q->bitrate;
  718. *data_size = 160 * sizeof(*outbuffer);
  719. return *data_size;
  720. }
  721. AVCodec qcelp_decoder =
  722. {
  723. .name = "qcelp",
  724. .type = CODEC_TYPE_AUDIO,
  725. .id = CODEC_ID_QCELP,
  726. .init = qcelp_decode_init,
  727. .decode = qcelp_decode_frame,
  728. .priv_data_size = sizeof(QCELPContext),
  729. .long_name = NULL_IF_CONFIG_SMALL("QCELP / PureVoice"),
  730. };