ra144enc.c
Go to the documentation of this file.
1 /*
2  * Real Audio 1.0 (14.4K) encoder
3  * Copyright (c) 2010 Francesco Lavra <francescolavra@interfree.it>
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 
28 #include <float.h>
29 
30 #include "avcodec.h"
31 #include "put_bits.h"
32 #include "celp_filters.h"
33 #include "ra144.h"
34 
35 
37 {
38  RA144Context *ractx;
39  int ret;
40 
41  if (avctx->sample_fmt != AV_SAMPLE_FMT_S16) {
42  av_log(avctx, AV_LOG_ERROR, "invalid sample format\n");
43  return -1;
44  }
45  if (avctx->channels != 1) {
46  av_log(avctx, AV_LOG_ERROR, "invalid number of channels: %d\n",
47  avctx->channels);
48  return -1;
49  }
50  avctx->frame_size = NBLOCKS * BLOCKSIZE;
51  avctx->bit_rate = 8000;
52  ractx = avctx->priv_data;
53  ractx->lpc_coef[0] = ractx->lpc_tables[0];
54  ractx->lpc_coef[1] = ractx->lpc_tables[1];
55  ractx->avctx = avctx;
56  ret = ff_lpc_init(&ractx->lpc_ctx, avctx->frame_size, LPC_ORDER,
58  return ret;
59 }
60 
61 
63 {
64  RA144Context *ractx = avctx->priv_data;
65  ff_lpc_end(&ractx->lpc_ctx);
66  return 0;
67 }
68 
69 
80 static int quantize(int value, const int16_t *table, unsigned int size)
81 {
82  unsigned int low = 0, high = size - 1;
83 
84  while (1) {
85  int index = (low + high) >> 1;
86  int error = table[index] - value;
87 
88  if (index == low)
89  return table[high] + error > value ? low : high;
90  if (error > 0) {
91  high = index;
92  } else {
93  low = index;
94  }
95  }
96 }
97 
98 
105 static void orthogonalize(float *v, const float *u)
106 {
107  int i;
108  float num = 0, den = 0;
109 
110  for (i = 0; i < BLOCKSIZE; i++) {
111  num += v[i] * u[i];
112  den += u[i] * u[i];
113  }
114  num /= den;
115  for (i = 0; i < BLOCKSIZE; i++)
116  v[i] -= num * u[i];
117 }
118 
119 
133 static void get_match_score(float *work, const float *coefs, float *vect,
134  const float *ortho1, const float *ortho2,
135  const float *data, float *score, float *gain)
136 {
137  float c, g;
138  int i;
139 
140  ff_celp_lp_synthesis_filterf(work, coefs, vect, BLOCKSIZE, LPC_ORDER);
141  if (ortho1)
142  orthogonalize(work, ortho1);
143  if (ortho2)
144  orthogonalize(work, ortho2);
145  c = g = 0;
146  for (i = 0; i < BLOCKSIZE; i++) {
147  g += work[i] * work[i];
148  c += data[i] * work[i];
149  }
150  if (c <= 0) {
151  *score = 0;
152  return;
153  }
154  *gain = c / g;
155  *score = *gain * c;
156 }
157 
158 
166 static void create_adapt_vect(float *vect, const int16_t *cb, int lag)
167 {
168  int i;
169 
170  cb += BUFFERSIZE - lag;
171  for (i = 0; i < FFMIN(BLOCKSIZE, lag); i++)
172  vect[i] = cb[i];
173  if (lag < BLOCKSIZE)
174  for (i = 0; i < BLOCKSIZE - lag; i++)
175  vect[lag + i] = cb[i];
176 }
177 
178 
189 static int adaptive_cb_search(const int16_t *adapt_cb, float *work,
190  const float *coefs, float *data)
191 {
192  int i, best_vect;
193  float score, gain, best_score, best_gain;
194  float exc[BLOCKSIZE];
195 
196  gain = best_score = 0;
197  for (i = BLOCKSIZE / 2; i <= BUFFERSIZE; i++) {
198  create_adapt_vect(exc, adapt_cb, i);
199  get_match_score(work, coefs, exc, NULL, NULL, data, &score, &gain);
200  if (score > best_score) {
201  best_score = score;
202  best_vect = i;
203  best_gain = gain;
204  }
205  }
206  if (!best_score)
207  return 0;
208 
213  create_adapt_vect(exc, adapt_cb, best_vect);
215  for (i = 0; i < BLOCKSIZE; i++)
216  data[i] -= best_gain * work[i];
217  return best_vect - BLOCKSIZE / 2 + 1;
218 }
219 
220 
237 static void find_best_vect(float *work, const float *coefs,
238  const int8_t cb[][BLOCKSIZE], const float *ortho1,
239  const float *ortho2, float *data, int *idx,
240  float *gain)
241 {
242  int i, j;
243  float g, score, best_score;
244  float vect[BLOCKSIZE];
245 
246  *idx = *gain = best_score = 0;
247  for (i = 0; i < FIXED_CB_SIZE; i++) {
248  for (j = 0; j < BLOCKSIZE; j++)
249  vect[j] = cb[i][j];
250  get_match_score(work, coefs, vect, ortho1, ortho2, data, &score, &g);
251  if (score > best_score) {
252  best_score = score;
253  *idx = i;
254  *gain = g;
255  }
256  }
257 }
258 
259 
272 static void fixed_cb_search(float *work, const float *coefs, float *data,
273  int cba_idx, int *cb1_idx, int *cb2_idx)
274 {
275  int i, ortho_cb1;
276  float gain;
277  float cba_vect[BLOCKSIZE], cb1_vect[BLOCKSIZE];
278  float vect[BLOCKSIZE];
279 
284  if (cba_idx)
285  memcpy(cba_vect, work, sizeof(cba_vect));
286 
287  find_best_vect(work, coefs, ff_cb1_vects, cba_idx ? cba_vect : NULL, NULL,
288  data, cb1_idx, &gain);
289 
294  if (gain) {
295  for (i = 0; i < BLOCKSIZE; i++)
296  vect[i] = ff_cb1_vects[*cb1_idx][i];
297  ff_celp_lp_synthesis_filterf(work, coefs, vect, BLOCKSIZE, LPC_ORDER);
298  if (cba_idx)
299  orthogonalize(work, cba_vect);
300  for (i = 0; i < BLOCKSIZE; i++)
301  data[i] -= gain * work[i];
302  memcpy(cb1_vect, work, sizeof(cb1_vect));
303  ortho_cb1 = 1;
304  } else
305  ortho_cb1 = 0;
306 
307  find_best_vect(work, coefs, ff_cb2_vects, cba_idx ? cba_vect : NULL,
308  ortho_cb1 ? cb1_vect : NULL, data, cb2_idx, &gain);
309 }
310 
311 
322  const int16_t *sblock_data,
323  const int16_t *lpc_coefs, unsigned int rms,
324  PutBitContext *pb)
325 {
326  float data[BLOCKSIZE], work[LPC_ORDER + BLOCKSIZE];
327  float coefs[LPC_ORDER];
328  float zero[BLOCKSIZE], cba[BLOCKSIZE], cb1[BLOCKSIZE], cb2[BLOCKSIZE];
329  int16_t cba_vect[BLOCKSIZE];
330  int cba_idx, cb1_idx, cb2_idx, gain;
331  int i, n, m[3];
332  float g[3];
333  float error, best_error;
334 
335  for (i = 0; i < LPC_ORDER; i++) {
336  work[i] = ractx->curr_sblock[BLOCKSIZE + i];
337  coefs[i] = lpc_coefs[i] * (1/4096.0);
338  }
339 
344  memset(data, 0, sizeof(data));
345  ff_celp_lp_synthesis_filterf(work + LPC_ORDER, coefs, data, BLOCKSIZE,
346  LPC_ORDER);
347  for (i = 0; i < BLOCKSIZE; i++) {
348  zero[i] = work[LPC_ORDER + i];
349  data[i] = sblock_data[i] - zero[i];
350  }
351 
357  memset(work, 0, LPC_ORDER * sizeof(*work));
358 
359  cba_idx = adaptive_cb_search(ractx->adapt_cb, work + LPC_ORDER, coefs,
360  data);
361  if (cba_idx) {
366  memcpy(cba, work + LPC_ORDER, sizeof(cba));
367 
368  ff_copy_and_dup(cba_vect, ractx->adapt_cb, cba_idx + BLOCKSIZE / 2 - 1);
369  m[0] = (ff_irms(cba_vect) * rms) >> 12;
370  }
371  fixed_cb_search(work + LPC_ORDER, coefs, data, cba_idx, &cb1_idx, &cb2_idx);
372  for (i = 0; i < BLOCKSIZE; i++) {
373  cb1[i] = ff_cb1_vects[cb1_idx][i];
374  cb2[i] = ff_cb2_vects[cb2_idx][i];
375  }
376  ff_celp_lp_synthesis_filterf(work + LPC_ORDER, coefs, cb1, BLOCKSIZE,
377  LPC_ORDER);
378  memcpy(cb1, work + LPC_ORDER, sizeof(cb1));
379  m[1] = (ff_cb1_base[cb1_idx] * rms) >> 8;
380  ff_celp_lp_synthesis_filterf(work + LPC_ORDER, coefs, cb2, BLOCKSIZE,
381  LPC_ORDER);
382  memcpy(cb2, work + LPC_ORDER, sizeof(cb2));
383  m[2] = (ff_cb2_base[cb2_idx] * rms) >> 8;
384  best_error = FLT_MAX;
385  gain = 0;
386  for (n = 0; n < 256; n++) {
387  g[1] = ((ff_gain_val_tab[n][1] * m[1]) >> ff_gain_exp_tab[n]) *
388  (1/4096.0);
389  g[2] = ((ff_gain_val_tab[n][2] * m[2]) >> ff_gain_exp_tab[n]) *
390  (1/4096.0);
391  error = 0;
392  if (cba_idx) {
393  g[0] = ((ff_gain_val_tab[n][0] * m[0]) >> ff_gain_exp_tab[n]) *
394  (1/4096.0);
395  for (i = 0; i < BLOCKSIZE; i++) {
396  data[i] = zero[i] + g[0] * cba[i] + g[1] * cb1[i] +
397  g[2] * cb2[i];
398  error += (data[i] - sblock_data[i]) *
399  (data[i] - sblock_data[i]);
400  }
401  } else {
402  for (i = 0; i < BLOCKSIZE; i++) {
403  data[i] = zero[i] + g[1] * cb1[i] + g[2] * cb2[i];
404  error += (data[i] - sblock_data[i]) *
405  (data[i] - sblock_data[i]);
406  }
407  }
408  if (error < best_error) {
409  best_error = error;
410  gain = n;
411  }
412  }
413  put_bits(pb, 7, cba_idx);
414  put_bits(pb, 8, gain);
415  put_bits(pb, 7, cb1_idx);
416  put_bits(pb, 7, cb2_idx);
417  ff_subblock_synthesis(ractx, lpc_coefs, cba_idx, cb1_idx, cb2_idx, rms,
418  gain);
419 }
420 
421 
422 static int ra144_encode_frame(AVCodecContext *avctx, uint8_t *frame,
423  int buf_size, void *data)
424 {
425  static const uint8_t sizes[LPC_ORDER] = {64, 32, 32, 16, 16, 8, 8, 8, 8, 4};
426  static const uint8_t bit_sizes[LPC_ORDER] = {6, 5, 5, 4, 4, 3, 3, 3, 3, 2};
427  RA144Context *ractx;
428  PutBitContext pb;
429  int32_t lpc_data[NBLOCKS * BLOCKSIZE];
430  int32_t lpc_coefs[LPC_ORDER][MAX_LPC_ORDER];
431  int shift[LPC_ORDER];
432  int16_t block_coefs[NBLOCKS][LPC_ORDER];
433  int lpc_refl[LPC_ORDER];
434  unsigned int refl_rms[NBLOCKS];
435  int energy = 0;
436  int i, idx;
437 
438  if (buf_size < FRAMESIZE) {
439  av_log(avctx, AV_LOG_ERROR, "output buffer too small\n");
440  return 0;
441  }
442  ractx = avctx->priv_data;
443 
451  for (i = 0; i < (2 * BLOCKSIZE + BLOCKSIZE / 2); i++) {
452  lpc_data[i] = ractx->curr_block[BLOCKSIZE + BLOCKSIZE / 2 + i];
453  energy += (lpc_data[i] * lpc_data[i]) >> 4;
454  }
455  for (i = 2 * BLOCKSIZE + BLOCKSIZE / 2; i < NBLOCKS * BLOCKSIZE; i++) {
456  lpc_data[i] = *((int16_t *)data + i - 2 * BLOCKSIZE - BLOCKSIZE / 2) >>
457  2;
458  energy += (lpc_data[i] * lpc_data[i]) >> 4;
459  }
460  energy = ff_energy_tab[quantize(ff_t_sqrt(energy >> 5) >> 10, ff_energy_tab,
461  32)];
462 
463  ff_lpc_calc_coefs(&ractx->lpc_ctx, lpc_data, NBLOCKS * BLOCKSIZE, LPC_ORDER,
464  LPC_ORDER, 16, lpc_coefs, shift, FF_LPC_TYPE_LEVINSON,
465  0, ORDER_METHOD_EST, 12, 0);
466  for (i = 0; i < LPC_ORDER; i++)
467  block_coefs[NBLOCKS - 1][i] = -(lpc_coefs[LPC_ORDER - 1][i] <<
468  (12 - shift[LPC_ORDER - 1]));
469 
475  if (ff_eval_refl(lpc_refl, block_coefs[NBLOCKS - 1], avctx)) {
479  ff_int_to_int16(block_coefs[NBLOCKS - 1], ractx->lpc_coef[1]);
480  if (ff_eval_refl(lpc_refl, block_coefs[NBLOCKS - 1], avctx)) {
481  /* the filter is still unstable. set reflection coeffs to zero. */
482  memset(lpc_refl, 0, sizeof(lpc_refl));
483  }
484  }
485  init_put_bits(&pb, frame, buf_size);
486  for (i = 0; i < LPC_ORDER; i++) {
487  idx = quantize(lpc_refl[i], ff_lpc_refl_cb[i], sizes[i]);
488  put_bits(&pb, bit_sizes[i], idx);
489  lpc_refl[i] = ff_lpc_refl_cb[i][idx];
490  }
491  ractx->lpc_refl_rms[0] = ff_rms(lpc_refl);
492  ff_eval_coefs(ractx->lpc_coef[0], lpc_refl);
493  refl_rms[0] = ff_interp(ractx, block_coefs[0], 1, 1, ractx->old_energy);
494  refl_rms[1] = ff_interp(ractx, block_coefs[1], 2,
495  energy <= ractx->old_energy,
496  ff_t_sqrt(energy * ractx->old_energy) >> 12);
497  refl_rms[2] = ff_interp(ractx, block_coefs[2], 3, 0, energy);
498  refl_rms[3] = ff_rescale_rms(ractx->lpc_refl_rms[0], energy);
499  ff_int_to_int16(block_coefs[NBLOCKS - 1], ractx->lpc_coef[0]);
500  put_bits(&pb, 5, quantize(energy, ff_energy_tab, 32));
501  for (i = 0; i < NBLOCKS; i++)
502  ra144_encode_subblock(ractx, ractx->curr_block + i * BLOCKSIZE,
503  block_coefs[i], refl_rms[i], &pb);
504  flush_put_bits(&pb);
505  ractx->old_energy = energy;
506  ractx->lpc_refl_rms[1] = ractx->lpc_refl_rms[0];
507  FFSWAP(unsigned int *, ractx->lpc_coef[0], ractx->lpc_coef[1]);
508  for (i = 0; i < NBLOCKS * BLOCKSIZE; i++)
509  ractx->curr_block[i] = *((int16_t *)data + i) >> 2;
510  return FRAMESIZE;
511 }
512 
513 
515  .name = "real_144",
516  .type = AVMEDIA_TYPE_AUDIO,
517  .id = CODEC_ID_RA_144,
518  .priv_data_size = sizeof(RA144Context),
520  .encode = ra144_encode_frame,
522  .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
524  .long_name = NULL_IF_CONFIG_SMALL("RealAudio 1.0 (14.4K) encoder"),
525 };