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.

794 lines
24KB

  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 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 "bitstream.h"
  31. #include "qcelpdata.h"
  32. #include "celp_math.h"
  33. #include "celp_filters.h"
  34. #undef NDEBUG
  35. #include <assert.h>
  36. typedef enum
  37. {
  38. I_F_Q = -1, /*!< insufficient frame quality */
  39. SILENCE,
  40. RATE_OCTAVE,
  41. RATE_QUARTER,
  42. RATE_HALF,
  43. RATE_FULL
  44. } qcelp_packet_rate;
  45. typedef struct
  46. {
  47. GetBitContext gb;
  48. qcelp_packet_rate bitrate;
  49. QCELPFrame frame; /*!< unpacked data frame */
  50. uint8_t erasure_count;
  51. uint8_t octave_count; /*!< count the consecutive RATE_OCTAVE frames */
  52. float prev_lspf[10];
  53. float predictor_lspf[10];/*!< LSP predictor for RATE_OCTAVE and I_F_Q */
  54. float pitch_synthesis_filter_mem[303];
  55. float pitch_pre_filter_mem[303];
  56. float rnd_fir_filter_mem[180];
  57. float formant_mem[170];
  58. float last_codebook_gain;
  59. int prev_g1[2];
  60. int prev_bitrate;
  61. float pitch_gain[4];
  62. uint8_t pitch_lag[4];
  63. uint16_t first16bits;
  64. } QCELPContext;
  65. /**
  66. * Reconstructs LPC coefficients from the line spectral pair frequencies.
  67. *
  68. * TIA/EIA/IS-733 2.4.3.3.5
  69. */
  70. void ff_qcelp_lspf2lpc(const float *lspf, float *lpc);
  71. static void weighted_vector_sumf(float *out, const float *in_a,
  72. const float *in_b, float weight_coeff_a,
  73. float weight_coeff_b, int length)
  74. {
  75. int i;
  76. for(i=0; i<length; i++)
  77. out[i] = weight_coeff_a * in_a[i]
  78. + weight_coeff_b * in_b[i];
  79. }
  80. /**
  81. * Initialize the speech codec according to the specification.
  82. *
  83. * TIA/EIA/IS-733 2.4.9
  84. */
  85. static av_cold int qcelp_decode_init(AVCodecContext *avctx)
  86. {
  87. QCELPContext *q = avctx->priv_data;
  88. int i;
  89. avctx->sample_fmt = SAMPLE_FMT_FLT;
  90. for(i=0; i<10; i++)
  91. q->prev_lspf[i] = (i+1)/11.;
  92. return 0;
  93. }
  94. /**
  95. * Decodes the 10 quantized LSP frequencies from the LSPV/LSP
  96. * transmission codes of any bitrate and checks for badly received packets.
  97. *
  98. * @param q the context
  99. * @param lspf line spectral pair frequencies
  100. *
  101. * @return 0 on success, -1 if the packet is badly received
  102. *
  103. * TIA/EIA/IS-733 2.4.3.2.6.2-2, 2.4.8.7.3
  104. */
  105. static int decode_lspf(QCELPContext *q, float *lspf)
  106. {
  107. int i;
  108. float tmp_lspf, smooth, erasure_coeff;
  109. const float *predictors;
  110. if(q->bitrate == RATE_OCTAVE || q->bitrate == I_F_Q)
  111. {
  112. predictors = (q->prev_bitrate != RATE_OCTAVE &&
  113. q->prev_bitrate != I_F_Q ?
  114. q->prev_lspf : q->predictor_lspf);
  115. if(q->bitrate == RATE_OCTAVE)
  116. {
  117. q->octave_count++;
  118. for(i=0; i<10; i++)
  119. {
  120. q->predictor_lspf[i] =
  121. lspf[i] = (q->frame.lspv[i] ? QCELP_LSP_SPREAD_FACTOR
  122. : -QCELP_LSP_SPREAD_FACTOR)
  123. + predictors[i] * QCELP_LSP_OCTAVE_PREDICTOR
  124. + (i + 1) * ((1 - QCELP_LSP_OCTAVE_PREDICTOR)/11);
  125. }
  126. smooth = (q->octave_count < 10 ? .875 : 0.1);
  127. }else
  128. {
  129. erasure_coeff = QCELP_LSP_OCTAVE_PREDICTOR;
  130. assert(q->bitrate == I_F_Q);
  131. if(q->erasure_count > 1)
  132. erasure_coeff *= (q->erasure_count < 4 ? 0.9 : 0.7);
  133. for(i=0; i<10; i++)
  134. {
  135. q->predictor_lspf[i] =
  136. lspf[i] = (i + 1) * ( 1 - erasure_coeff)/11
  137. + erasure_coeff * predictors[i];
  138. }
  139. smooth = 0.125;
  140. }
  141. // Check the stability of the LSP frequencies.
  142. lspf[0] = FFMAX(lspf[0], QCELP_LSP_SPREAD_FACTOR);
  143. for(i=1; i<10; i++)
  144. lspf[i] = FFMAX(lspf[i], (lspf[i-1] + QCELP_LSP_SPREAD_FACTOR));
  145. lspf[9] = FFMIN(lspf[9], (1.0 - QCELP_LSP_SPREAD_FACTOR));
  146. for(i=9; i>0; i--)
  147. lspf[i-1] = FFMIN(lspf[i-1], (lspf[i] - QCELP_LSP_SPREAD_FACTOR));
  148. // Low-pass filter the LSP frequencies.
  149. weighted_vector_sumf(lspf, lspf, q->prev_lspf, smooth, 1.0-smooth, 10);
  150. }else
  151. {
  152. q->octave_count = 0;
  153. tmp_lspf = 0.;
  154. for(i=0; i<5 ; i++)
  155. {
  156. lspf[2*i+0] = tmp_lspf += qcelp_lspvq[i][q->frame.lspv[i]][0] * 0.0001;
  157. lspf[2*i+1] = tmp_lspf += qcelp_lspvq[i][q->frame.lspv[i]][1] * 0.0001;
  158. }
  159. // Check for badly received packets.
  160. if(q->bitrate == RATE_QUARTER)
  161. {
  162. if(lspf[9] <= .70 || lspf[9] >= .97)
  163. return -1;
  164. for(i=3; i<10; i++)
  165. if(fabs(lspf[i] - lspf[i-2]) < .08)
  166. return -1;
  167. }else
  168. {
  169. if(lspf[9] <= .66 || lspf[9] >= .985)
  170. return -1;
  171. for(i=4; i<10; i++)
  172. if (fabs(lspf[i] - lspf[i-4]) < .0931)
  173. return -1;
  174. }
  175. }
  176. return 0;
  177. }
  178. /**
  179. * Converts codebook transmission codes to GAIN and INDEX.
  180. *
  181. * @param q the context
  182. * @param gain array holding the decoded gain
  183. *
  184. * TIA/EIA/IS-733 2.4.6.2
  185. */
  186. static void decode_gain_and_index(QCELPContext *q,
  187. float *gain) {
  188. int i, subframes_count, g1[16];
  189. float slope;
  190. if(q->bitrate >= RATE_QUARTER)
  191. {
  192. switch(q->bitrate)
  193. {
  194. case RATE_FULL: subframes_count = 16; break;
  195. case RATE_HALF: subframes_count = 4; break;
  196. default: subframes_count = 5;
  197. }
  198. for(i=0; i<subframes_count; i++)
  199. {
  200. g1[i] = 4 * q->frame.cbgain[i];
  201. if(q->bitrate == RATE_FULL && !((i+1) & 3))
  202. {
  203. g1[i] += av_clip((g1[i-1] + g1[i-2] + g1[i-3]) / 3 - 6, 0, 32);
  204. }
  205. gain[i] = qcelp_g12ga[g1[i]];
  206. if(q->frame.cbsign[i])
  207. {
  208. gain[i] = -gain[i];
  209. q->frame.cindex[i] = (q->frame.cindex[i]-89) & 127;
  210. }
  211. }
  212. q->prev_g1[0] = g1[i-2];
  213. q->prev_g1[1] = g1[i-1];
  214. q->last_codebook_gain = qcelp_g12ga[g1[i-1]];
  215. if(q->bitrate == RATE_QUARTER)
  216. {
  217. // Provide smoothing of the unvoiced excitation energy.
  218. gain[7] = gain[4];
  219. gain[6] = 0.4*gain[3] + 0.6*gain[4];
  220. gain[5] = gain[3];
  221. gain[4] = 0.8*gain[2] + 0.2*gain[3];
  222. gain[3] = 0.2*gain[1] + 0.8*gain[2];
  223. gain[2] = gain[1];
  224. gain[1] = 0.6*gain[0] + 0.4*gain[1];
  225. }
  226. }else
  227. {
  228. if(q->bitrate == RATE_OCTAVE)
  229. {
  230. g1[0] = 2 * q->frame.cbgain[0]
  231. + av_clip((q->prev_g1[0] + q->prev_g1[1]) / 2 - 5, 0, 54);
  232. subframes_count = 8;
  233. }else
  234. {
  235. assert(q->bitrate == I_F_Q);
  236. g1[0] = q->prev_g1[1];
  237. switch(q->erasure_count)
  238. {
  239. case 1 : break;
  240. case 2 : g1[0] -= 1; break;
  241. case 3 : g1[0] -= 2; break;
  242. default: g1[0] -= 6;
  243. }
  244. if(g1[0] < 0)
  245. g1[0] = 0;
  246. subframes_count = 4;
  247. }
  248. // This interpolation is done to produce smoother background noise.
  249. slope = 0.5*(qcelp_g12ga[g1[0]] - q->last_codebook_gain) / subframes_count;
  250. for(i=1; i<=subframes_count; i++)
  251. gain[i-1] = q->last_codebook_gain + slope * i;
  252. q->last_codebook_gain = gain[i-2];
  253. q->prev_g1[0] = q->prev_g1[1];
  254. q->prev_g1[1] = g1[0];
  255. }
  256. }
  257. /**
  258. * If the received packet is Rate 1/4 a further sanity check is made of the
  259. * codebook gain.
  260. *
  261. * @param cbgain the unpacked cbgain array
  262. * @return -1 if the sanity check fails, 0 otherwise
  263. *
  264. * TIA/EIA/IS-733 2.4.8.7.3
  265. */
  266. static int codebook_sanity_check_for_rate_quarter(const uint8_t *cbgain)
  267. {
  268. int i, diff, prev_diff=0;
  269. for(i=1; i<5; i++)
  270. {
  271. diff = cbgain[i] - cbgain[i-1];
  272. if(FFABS(diff) > 10)
  273. return -1;
  274. else if(FFABS(diff - prev_diff) > 12)
  275. return -1;
  276. prev_diff = diff;
  277. }
  278. return 0;
  279. }
  280. /**
  281. * Computes the scaled codebook vector Cdn From INDEX and GAIN
  282. * for all rates.
  283. *
  284. * The specification lacks some information here.
  285. *
  286. * TIA/EIA/IS-733 has an omission on the codebook index determination
  287. * formula for RATE_FULL and RATE_HALF frames at section 2.4.8.1.1. It says
  288. * you have to subtract the decoded index parameter from the given scaled
  289. * codebook vector index 'n' to get the desired circular codebook index, but
  290. * it does not mention that you have to clamp 'n' to [0-9] in order to get
  291. * RI-compliant results.
  292. *
  293. * The reason for this mistake seems to be the fact they forgot to mention you
  294. * have to do these calculations per codebook subframe and adjust given
  295. * equation values accordingly.
  296. *
  297. * @param q the context
  298. * @param gain array holding the 4 pitch subframe gain values
  299. * @param cdn_vector array for the generated scaled codebook vector
  300. */
  301. static void compute_svector(QCELPContext *q, const float *gain,
  302. float *cdn_vector)
  303. {
  304. int i, j, k;
  305. uint16_t cbseed, cindex;
  306. float *rnd, tmp_gain, fir_filter_value;
  307. switch(q->bitrate)
  308. {
  309. case RATE_FULL:
  310. for(i=0; i<16; i++)
  311. {
  312. tmp_gain = gain[i] * QCELP_RATE_FULL_CODEBOOK_RATIO;
  313. cindex = -q->frame.cindex[i];
  314. for(j=0; j<10; j++)
  315. *cdn_vector++ = tmp_gain * qcelp_rate_full_codebook[cindex++ & 127];
  316. }
  317. break;
  318. case RATE_HALF:
  319. for(i=0; i<4; i++)
  320. {
  321. tmp_gain = gain[i] * QCELP_RATE_HALF_CODEBOOK_RATIO;
  322. cindex = -q->frame.cindex[i];
  323. for (j = 0; j < 40; j++)
  324. *cdn_vector++ = tmp_gain * qcelp_rate_half_codebook[cindex++ & 127];
  325. }
  326. break;
  327. case RATE_QUARTER:
  328. cbseed = (0x0003 & q->frame.lspv[4])<<14 |
  329. (0x003F & q->frame.lspv[3])<< 8 |
  330. (0x0060 & q->frame.lspv[2])<< 1 |
  331. (0x0007 & q->frame.lspv[1])<< 3 |
  332. (0x0038 & q->frame.lspv[0])>> 3 ;
  333. rnd = q->rnd_fir_filter_mem + 20;
  334. for(i=0; i<8; i++)
  335. {
  336. tmp_gain = gain[i] * (QCELP_SQRT1887 / 32768.0);
  337. for(k=0; k<20; k++)
  338. {
  339. cbseed = 521 * cbseed + 259;
  340. *rnd = (int16_t)cbseed;
  341. // FIR filter
  342. fir_filter_value = 0.0;
  343. for(j=0; j<10; j++)
  344. fir_filter_value += qcelp_rnd_fir_coefs[j ]
  345. * (rnd[-j ] + rnd[-20+j]);
  346. fir_filter_value += qcelp_rnd_fir_coefs[10] * rnd[-10];
  347. *cdn_vector++ = tmp_gain * fir_filter_value;
  348. rnd++;
  349. }
  350. }
  351. memcpy(q->rnd_fir_filter_mem, q->rnd_fir_filter_mem + 160, 20 * sizeof(float));
  352. break;
  353. case RATE_OCTAVE:
  354. cbseed = q->first16bits;
  355. for(i=0; i<8; i++)
  356. {
  357. tmp_gain = gain[i] * (QCELP_SQRT1887 / 32768.0);
  358. for(j=0; j<20; j++)
  359. {
  360. cbseed = 521 * cbseed + 259;
  361. *cdn_vector++ = tmp_gain * (int16_t)cbseed;
  362. }
  363. }
  364. break;
  365. case I_F_Q:
  366. cbseed = -44; // random codebook index
  367. for(i=0; i<4; i++)
  368. {
  369. tmp_gain = gain[i] * QCELP_RATE_FULL_CODEBOOK_RATIO;
  370. for(j=0; j<40; j++)
  371. *cdn_vector++ = tmp_gain * qcelp_rate_full_codebook[cbseed++ & 127];
  372. }
  373. break;
  374. }
  375. }
  376. /**
  377. * Apply generic gain control.
  378. *
  379. * @param v_out output vector
  380. * @param v_in gain-controlled vector
  381. * @param v_ref vector to control gain of
  382. *
  383. * FIXME: If v_ref is a zero vector, it energy is zero
  384. * and the behavior of the gain control is
  385. * undefined in the specs.
  386. *
  387. * TIA/EIA/IS-733 2.4.8.3-2/3/4/5, 2.4.8.6
  388. */
  389. static void apply_gain_ctrl(float *v_out, const float *v_ref,
  390. const float *v_in)
  391. {
  392. int i, j, len;
  393. float scalefactor;
  394. for(i=0, j=0; i<4; i++)
  395. {
  396. scalefactor = ff_dot_productf(v_in + j, v_in + j, 40);
  397. if(scalefactor)
  398. scalefactor = sqrt(ff_dot_productf(v_ref + j, v_ref + j, 40)
  399. / scalefactor);
  400. else
  401. av_log_missing_feature(NULL, "Zero energy for gain control", 1);
  402. for(len=j+40; j<len; j++)
  403. v_out[j] = scalefactor * v_in[j];
  404. }
  405. }
  406. /**
  407. * Apply filter in pitch-subframe steps.
  408. *
  409. * @param memory buffer for the previous state of the filter
  410. * - must be able to contain 303 elements
  411. * - the 143 first elements are from the previous state
  412. * - the next 160 are for output
  413. * @param v_in input filter vector
  414. * @param gain per-subframe gain array, each element is between 0.0 and 2.0
  415. * @param lag per-subframe lag array, each element is
  416. * - between 16 and 143 if its corresponding pfrac is 0,
  417. * - between 16 and 139 otherwise
  418. * @param pfrac per-subframe boolean array, 1 if the lag is fractional, 0
  419. * otherwise
  420. *
  421. * @return filter output vector
  422. */
  423. static const float *do_pitchfilter(float memory[303], const float v_in[160],
  424. const float gain[4], const uint8_t *lag,
  425. const uint8_t pfrac[4])
  426. {
  427. int i, j;
  428. float *v_lag, *v_out;
  429. const float *v_len;
  430. v_out = memory + 143; // Output vector starts at memory[143].
  431. for(i=0; i<4; i++)
  432. {
  433. if(gain[i])
  434. {
  435. v_lag = memory + 143 + 40 * i - lag[i];
  436. for(v_len=v_in+40; v_in<v_len; v_in++)
  437. {
  438. if(pfrac[i]) // If it is a fractional lag...
  439. {
  440. for(j=0, *v_out=0.; j<4; j++)
  441. *v_out += qcelp_hammsinc_table[j] * (v_lag[j-4] + v_lag[3-j]);
  442. }else
  443. *v_out = *v_lag;
  444. *v_out = *v_in + gain[i] * *v_out;
  445. v_lag++;
  446. v_out++;
  447. }
  448. }else
  449. {
  450. memcpy(v_out, v_in, 40 * sizeof(float));
  451. v_in += 40;
  452. v_out += 40;
  453. }
  454. }
  455. memmove(memory, memory + 160, 143 * sizeof(float));
  456. return memory + 143;
  457. }
  458. /**
  459. * Apply pitch synthesis filter and pitch prefilter to the scaled codebook vector.
  460. * TIA/EIA/IS-733 2.4.5.2
  461. *
  462. * @param q the context
  463. * @param cdn_vector the scaled codebook vector
  464. */
  465. static void apply_pitch_filters(QCELPContext *q, float *cdn_vector)
  466. {
  467. int i;
  468. const float *v_synthesis_filtered, *v_pre_filtered;
  469. if(q->bitrate >= RATE_HALF ||
  470. (q->bitrate == I_F_Q && (q->prev_bitrate >= RATE_HALF)))
  471. {
  472. if(q->bitrate >= RATE_HALF)
  473. {
  474. // Compute gain & lag for the whole frame.
  475. for(i=0; i<4; i++)
  476. {
  477. q->pitch_gain[i] = q->frame.plag[i] ? (q->frame.pgain[i] + 1) * 0.25 : 0.0;
  478. q->pitch_lag[i] = q->frame.plag[i] + 16;
  479. }
  480. }else
  481. {
  482. float max_pitch_gain = q->erasure_count < 3 ? 0.9 - 0.3 * (q->erasure_count - 1) : 0.0;
  483. for(i=0; i<4; i++)
  484. q->pitch_gain[i] = FFMIN(q->pitch_gain[i], max_pitch_gain);
  485. memset(q->frame.pfrac, 0, sizeof(q->frame.pfrac));
  486. }
  487. // pitch synthesis filter
  488. v_synthesis_filtered = do_pitchfilter(q->pitch_synthesis_filter_mem,
  489. cdn_vector, q->pitch_gain,
  490. q->pitch_lag, q->frame.pfrac);
  491. // pitch prefilter update
  492. for(i=0; i<4; i++)
  493. q->pitch_gain[i] = 0.5 * FFMIN(q->pitch_gain[i], 1.0);
  494. v_pre_filtered = do_pitchfilter(q->pitch_pre_filter_mem,
  495. v_synthesis_filtered,
  496. q->pitch_gain, q->pitch_lag,
  497. q->frame.pfrac);
  498. apply_gain_ctrl(cdn_vector, v_synthesis_filtered, v_pre_filtered);
  499. }else
  500. {
  501. memcpy(q->pitch_synthesis_filter_mem, cdn_vector + 17,
  502. 143 * sizeof(float));
  503. memcpy(q->pitch_pre_filter_mem, cdn_vector + 17, 143 * sizeof(float));
  504. memset(q->pitch_gain, 0, sizeof(q->pitch_gain));
  505. memset(q->pitch_lag, 0, sizeof(q->pitch_lag));
  506. }
  507. }
  508. /**
  509. * Interpolates LSP frequencies and computes LPC coefficients
  510. * for a given bitrate & pitch subframe.
  511. *
  512. * TIA/EIA/IS-733 2.4.3.3.4
  513. *
  514. * @param q the context
  515. * @param curr_lspf LSP frequencies vector of the current frame
  516. * @param lpc float vector for the resulting LPC
  517. * @param subframe_num frame number in decoded stream
  518. */
  519. void interpolate_lpc(QCELPContext *q, const float *curr_lspf, float *lpc,
  520. const int subframe_num)
  521. {
  522. float interpolated_lspf[10];
  523. float weight;
  524. if(q->bitrate >= RATE_QUARTER)
  525. weight = 0.25 * (subframe_num + 1);
  526. else if(q->bitrate == RATE_OCTAVE && !subframe_num)
  527. weight = 0.625;
  528. else
  529. weight = 1.0;
  530. if(weight != 1.0)
  531. {
  532. weighted_vector_sumf(interpolated_lspf, curr_lspf, q->prev_lspf,
  533. weight, 1.0 - weight, 10);
  534. ff_qcelp_lspf2lpc(interpolated_lspf, lpc);
  535. }else if(q->bitrate >= RATE_QUARTER ||
  536. (q->bitrate == I_F_Q && !subframe_num))
  537. ff_qcelp_lspf2lpc(curr_lspf, lpc);
  538. }
  539. static qcelp_packet_rate buf_size2bitrate(const int buf_size)
  540. {
  541. switch(buf_size)
  542. {
  543. case 35: return RATE_FULL;
  544. case 17: return RATE_HALF;
  545. case 8: return RATE_QUARTER;
  546. case 4: return RATE_OCTAVE;
  547. case 1: return SILENCE;
  548. }
  549. return I_F_Q;
  550. }
  551. /**
  552. * Determine the bitrate from the frame size and/or the first byte of the frame.
  553. *
  554. * @param avctx the AV codec context
  555. * @param buf_size length of the buffer
  556. * @param buf the bufffer
  557. *
  558. * @return the bitrate on success,
  559. * I_F_Q if the bitrate cannot be satisfactorily determined
  560. *
  561. * TIA/EIA/IS-733 2.4.8.7.1
  562. */
  563. static int determine_bitrate(AVCodecContext *avctx, const int buf_size,
  564. const uint8_t **buf)
  565. {
  566. qcelp_packet_rate bitrate;
  567. if((bitrate = buf_size2bitrate(buf_size)) >= 0)
  568. {
  569. if(bitrate > **buf)
  570. {
  571. av_log(avctx, AV_LOG_WARNING,
  572. "Claimed bitrate and buffer size mismatch.\n");
  573. bitrate = **buf;
  574. }else if(bitrate < **buf)
  575. {
  576. av_log(avctx, AV_LOG_ERROR,
  577. "Buffer is too small for the claimed bitrate.\n");
  578. return I_F_Q;
  579. }
  580. (*buf)++;
  581. }else if((bitrate = buf_size2bitrate(buf_size + 1)) >= 0)
  582. {
  583. av_log(avctx, AV_LOG_WARNING,
  584. "Bitrate byte is missing, guessing the bitrate from packet size.\n");
  585. }else
  586. return I_F_Q;
  587. if(bitrate == SILENCE)
  588. {
  589. // FIXME: the decoder should not handle SILENCE frames as I_F_Q frames
  590. av_log_missing_feature(avctx, "Blank frame", 1);
  591. bitrate = I_F_Q;
  592. }
  593. return bitrate;
  594. }
  595. static void warn_insufficient_frame_quality(AVCodecContext *avctx,
  596. const char *message)
  597. {
  598. av_log(avctx, AV_LOG_WARNING, "Frame #%d, IFQ: %s\n", avctx->frame_number,
  599. message);
  600. }
  601. static int qcelp_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
  602. const uint8_t *buf, int buf_size)
  603. {
  604. QCELPContext *q = avctx->priv_data;
  605. float *outbuffer = data;
  606. int i;
  607. float quantized_lspf[10], lpc[10];
  608. float gain[16];
  609. float *formant_mem;
  610. if((q->bitrate = determine_bitrate(avctx, buf_size, &buf)) == I_F_Q)
  611. {
  612. warn_insufficient_frame_quality(avctx, "bitrate cannot be determined.");
  613. goto erasure;
  614. }
  615. if(q->bitrate == RATE_OCTAVE &&
  616. (q->first16bits = AV_RB16(buf)) == 0xFFFF)
  617. {
  618. warn_insufficient_frame_quality(avctx, "Bitrate is 1/8 and first 16 bits are on.");
  619. goto erasure;
  620. }
  621. if(q->bitrate > SILENCE)
  622. {
  623. const QCELPBitmap *bitmaps = qcelp_unpacking_bitmaps_per_rate[q->bitrate];
  624. const QCELPBitmap *bitmaps_end = qcelp_unpacking_bitmaps_per_rate[q->bitrate]
  625. + qcelp_unpacking_bitmaps_lengths[q->bitrate];
  626. uint8_t *unpacked_data = (uint8_t *)&q->frame;
  627. init_get_bits(&q->gb, buf, 8*buf_size);
  628. memset(&q->frame, 0, sizeof(QCELPFrame));
  629. for(; bitmaps < bitmaps_end; bitmaps++)
  630. unpacked_data[bitmaps->index] |= get_bits(&q->gb, bitmaps->bitlen) << bitmaps->bitpos;
  631. // Check for erasures/blanks on rates 1, 1/4 and 1/8.
  632. if(q->frame.reserved)
  633. {
  634. warn_insufficient_frame_quality(avctx, "Wrong data in reserved frame area.");
  635. goto erasure;
  636. }
  637. if(q->bitrate == RATE_QUARTER &&
  638. codebook_sanity_check_for_rate_quarter(q->frame.cbgain))
  639. {
  640. warn_insufficient_frame_quality(avctx, "Codebook gain sanity check failed.");
  641. goto erasure;
  642. }
  643. if(q->bitrate >= RATE_HALF)
  644. {
  645. for(i=0; i<4; i++)
  646. {
  647. if(q->frame.pfrac[i] && q->frame.plag[i] >= 124)
  648. {
  649. warn_insufficient_frame_quality(avctx, "Cannot initialize pitch filter.");
  650. goto erasure;
  651. }
  652. }
  653. }
  654. }
  655. decode_gain_and_index(q, gain);
  656. compute_svector(q, gain, outbuffer);
  657. if(decode_lspf(q, quantized_lspf) < 0)
  658. {
  659. warn_insufficient_frame_quality(avctx, "Badly received packets in frame.");
  660. goto erasure;
  661. }
  662. apply_pitch_filters(q, outbuffer);
  663. if(q->bitrate == I_F_Q)
  664. {
  665. erasure:
  666. q->bitrate = I_F_Q;
  667. q->erasure_count++;
  668. decode_gain_and_index(q, gain);
  669. compute_svector(q, gain, outbuffer);
  670. decode_lspf(q, quantized_lspf);
  671. apply_pitch_filters(q, outbuffer);
  672. }else
  673. q->erasure_count = 0;
  674. formant_mem = q->formant_mem + 10;
  675. for(i=0; i<4; i++)
  676. {
  677. interpolate_lpc(q, quantized_lspf, lpc, i);
  678. ff_celp_lp_synthesis_filterf(formant_mem, lpc, outbuffer + i * 40, 40,
  679. 10);
  680. formant_mem += 40;
  681. }
  682. memcpy(q->formant_mem, q->formant_mem + 160, 10 * sizeof(float));
  683. // FIXME: postfilter and final gain control should be here.
  684. // TIA/EIA/IS-733 2.4.8.6
  685. formant_mem = q->formant_mem + 10;
  686. for(i=0; i<160; i++)
  687. *outbuffer++ = av_clipf(*formant_mem++, QCELP_CLIP_LOWER_BOUND,
  688. QCELP_CLIP_UPPER_BOUND);
  689. memcpy(q->prev_lspf, quantized_lspf, sizeof(q->prev_lspf));
  690. q->prev_bitrate = q->bitrate;
  691. *data_size = 160 * sizeof(*outbuffer);
  692. return *data_size;
  693. }
  694. AVCodec qcelp_decoder =
  695. {
  696. .name = "qcelp",
  697. .type = CODEC_TYPE_AUDIO,
  698. .id = CODEC_ID_QCELP,
  699. .init = qcelp_decode_init,
  700. .decode = qcelp_decode_frame,
  701. .priv_data_size = sizeof(QCELPContext),
  702. .long_name = NULL_IF_CONFIG_SMALL("QCELP / PureVoice"),
  703. };