g723_1.c
Go to the documentation of this file.
1 /*
2  * G.723.1 compatible decoder
3  * Copyright (c) 2006 Benjamin Larsson
4  * Copyright (c) 2010 Mohamed Naufal Basheer
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 
28 #define BITSTREAM_READER_LE
30 #include "libavutil/mem.h"
31 #include "libavutil/opt.h"
32 #include "avcodec.h"
33 #include "get_bits.h"
34 #include "acelp_vectors.h"
35 #include "celp_filters.h"
36 #include "g723_1_data.h"
37 #include "internal.h"
38 
39 #define CNG_RANDOM_SEED 12345
40 
44 enum FrameType {
48 };
49 
50 enum Rate {
53 };
54 
58 typedef struct {
59  int ad_cb_lag;
64  int amp_index;
65  int pulse_pos;
67 
71 typedef struct {
72  int index;
73  int16_t opt_gain;
74  int16_t sc_gain;
75 } PPFParam;
76 
77 typedef struct g723_1_context {
78  AVClass *class;
80 
84  enum Rate cur_rate;
86  int pitch_lag[2];
88 
89  int16_t prev_lsp[LPC_ORDER];
90  int16_t sid_lsp[LPC_ORDER];
92  int16_t excitation[PITCH_MAX + FRAME_LEN + 4];
93  int16_t synth_mem[LPC_ORDER];
94  int16_t fir_mem[LPC_ORDER];
96 
101  int sid_gain;
102  int cur_gain;
104  int pf_gain;
106 
107  int16_t audio[FRAME_LEN + LPC_ORDER + PITCH_MAX + 4];
109 
111 {
112  G723_1_Context *p = avctx->priv_data;
113 
115  avctx->sample_fmt = AV_SAMPLE_FMT_S16;
116  avctx->channels = 1;
117  avctx->sample_rate = 8000;
118  p->pf_gain = 1 << 12;
119 
121  avctx->coded_frame = &p->frame;
122 
123  memcpy(p->prev_lsp, dc_lsp, LPC_ORDER * sizeof(*p->prev_lsp));
124  memcpy(p->sid_lsp, dc_lsp, LPC_ORDER * sizeof(*p->sid_lsp));
125 
128 
129  return 0;
130 }
131 
139 static int unpack_bitstream(G723_1_Context *p, const uint8_t *buf,
140  int buf_size)
141 {
142  GetBitContext gb;
143  int ad_cb_len;
144  int temp, info_bits, i;
145 
146  init_get_bits(&gb, buf, buf_size * 8);
147 
148  /* Extract frame type and rate info */
149  info_bits = get_bits(&gb, 2);
150 
151  if (info_bits == 3) {
153  return 0;
154  }
155 
156  /* Extract 24 bit lsp indices, 8 bit for each band */
157  p->lsp_index[2] = get_bits(&gb, 8);
158  p->lsp_index[1] = get_bits(&gb, 8);
159  p->lsp_index[0] = get_bits(&gb, 8);
160 
161  if (info_bits == 2) {
163  p->subframe[0].amp_index = get_bits(&gb, 6);
164  return 0;
165  }
166 
167  /* Extract the info common to both rates */
168  p->cur_rate = info_bits ? RATE_5300 : RATE_6300;
170 
171  p->pitch_lag[0] = get_bits(&gb, 7);
172  if (p->pitch_lag[0] > 123) /* test if forbidden code */
173  return -1;
174  p->pitch_lag[0] += PITCH_MIN;
175  p->subframe[1].ad_cb_lag = get_bits(&gb, 2);
176 
177  p->pitch_lag[1] = get_bits(&gb, 7);
178  if (p->pitch_lag[1] > 123)
179  return -1;
180  p->pitch_lag[1] += PITCH_MIN;
181  p->subframe[3].ad_cb_lag = get_bits(&gb, 2);
182  p->subframe[0].ad_cb_lag = 1;
183  p->subframe[2].ad_cb_lag = 1;
184 
185  for (i = 0; i < SUBFRAMES; i++) {
186  /* Extract combined gain */
187  temp = get_bits(&gb, 12);
188  ad_cb_len = 170;
189  p->subframe[i].dirac_train = 0;
190  if (p->cur_rate == RATE_6300 && p->pitch_lag[i >> 1] < SUBFRAME_LEN - 2) {
191  p->subframe[i].dirac_train = temp >> 11;
192  temp &= 0x7FF;
193  ad_cb_len = 85;
194  }
195  p->subframe[i].ad_cb_gain = FASTDIV(temp, GAIN_LEVELS);
196  if (p->subframe[i].ad_cb_gain < ad_cb_len) {
197  p->subframe[i].amp_index = temp - p->subframe[i].ad_cb_gain *
198  GAIN_LEVELS;
199  } else {
200  return -1;
201  }
202  }
203 
204  p->subframe[0].grid_index = get_bits(&gb, 1);
205  p->subframe[1].grid_index = get_bits(&gb, 1);
206  p->subframe[2].grid_index = get_bits(&gb, 1);
207  p->subframe[3].grid_index = get_bits(&gb, 1);
208 
209  if (p->cur_rate == RATE_6300) {
210  skip_bits(&gb, 1); /* skip reserved bit */
211 
212  /* Compute pulse_pos index using the 13-bit combined position index */
213  temp = get_bits(&gb, 13);
214  p->subframe[0].pulse_pos = temp / 810;
215 
216  temp -= p->subframe[0].pulse_pos * 810;
217  p->subframe[1].pulse_pos = FASTDIV(temp, 90);
218 
219  temp -= p->subframe[1].pulse_pos * 90;
220  p->subframe[2].pulse_pos = FASTDIV(temp, 9);
221  p->subframe[3].pulse_pos = temp - p->subframe[2].pulse_pos * 9;
222 
223  p->subframe[0].pulse_pos = (p->subframe[0].pulse_pos << 16) +
224  get_bits(&gb, 16);
225  p->subframe[1].pulse_pos = (p->subframe[1].pulse_pos << 14) +
226  get_bits(&gb, 14);
227  p->subframe[2].pulse_pos = (p->subframe[2].pulse_pos << 16) +
228  get_bits(&gb, 16);
229  p->subframe[3].pulse_pos = (p->subframe[3].pulse_pos << 14) +
230  get_bits(&gb, 14);
231 
232  p->subframe[0].pulse_sign = get_bits(&gb, 6);
233  p->subframe[1].pulse_sign = get_bits(&gb, 5);
234  p->subframe[2].pulse_sign = get_bits(&gb, 6);
235  p->subframe[3].pulse_sign = get_bits(&gb, 5);
236  } else { /* 5300 bps */
237  p->subframe[0].pulse_pos = get_bits(&gb, 12);
238  p->subframe[1].pulse_pos = get_bits(&gb, 12);
239  p->subframe[2].pulse_pos = get_bits(&gb, 12);
240  p->subframe[3].pulse_pos = get_bits(&gb, 12);
241 
242  p->subframe[0].pulse_sign = get_bits(&gb, 4);
243  p->subframe[1].pulse_sign = get_bits(&gb, 4);
244  p->subframe[2].pulse_sign = get_bits(&gb, 4);
245  p->subframe[3].pulse_sign = get_bits(&gb, 4);
246  }
247 
248  return 0;
249 }
250 
254 static int16_t square_root(int val)
255 {
256  int16_t res = 0;
257  int16_t exp = 0x4000;
258  int i;
259 
260  for (i = 0; i < 14; i ++) {
261  int res_exp = res + exp;
262  if (val >= res_exp * res_exp << 1)
263  res += exp;
264  exp >>= 1;
265  }
266  return res;
267 }
268 
275 static int normalize_bits(int num, int width)
276 {
277  return width - av_log2(num) - 1;
278 }
279 
283 static int scale_vector(int16_t *dst, const int16_t *vector, int length)
284 {
285  int bits, max = 0;
286  int i;
287 
288 
289  for (i = 0; i < length; i++)
290  max |= FFABS(vector[i]);
291 
292  max = FFMIN(max, 0x7FFF);
293  bits = normalize_bits(max, 15);
294 
295  for (i = 0; i < length; i++)
296  dst[i] = vector[i] << bits >> 3;
297 
298  return bits - 3;
299 }
300 
309 static void inverse_quant(int16_t *cur_lsp, int16_t *prev_lsp,
310  uint8_t *lsp_index, int bad_frame)
311 {
312  int min_dist, pred;
313  int i, j, temp, stable;
314 
315  /* Check for frame erasure */
316  if (!bad_frame) {
317  min_dist = 0x100;
318  pred = 12288;
319  } else {
320  min_dist = 0x200;
321  pred = 23552;
322  lsp_index[0] = lsp_index[1] = lsp_index[2] = 0;
323  }
324 
325  /* Get the VQ table entry corresponding to the transmitted index */
326  cur_lsp[0] = lsp_band0[lsp_index[0]][0];
327  cur_lsp[1] = lsp_band0[lsp_index[0]][1];
328  cur_lsp[2] = lsp_band0[lsp_index[0]][2];
329  cur_lsp[3] = lsp_band1[lsp_index[1]][0];
330  cur_lsp[4] = lsp_band1[lsp_index[1]][1];
331  cur_lsp[5] = lsp_band1[lsp_index[1]][2];
332  cur_lsp[6] = lsp_band2[lsp_index[2]][0];
333  cur_lsp[7] = lsp_band2[lsp_index[2]][1];
334  cur_lsp[8] = lsp_band2[lsp_index[2]][2];
335  cur_lsp[9] = lsp_band2[lsp_index[2]][3];
336 
337  /* Add predicted vector & DC component to the previously quantized vector */
338  for (i = 0; i < LPC_ORDER; i++) {
339  temp = ((prev_lsp[i] - dc_lsp[i]) * pred + (1 << 14)) >> 15;
340  cur_lsp[i] += dc_lsp[i] + temp;
341  }
342 
343  for (i = 0; i < LPC_ORDER; i++) {
344  cur_lsp[0] = FFMAX(cur_lsp[0], 0x180);
345  cur_lsp[LPC_ORDER - 1] = FFMIN(cur_lsp[LPC_ORDER - 1], 0x7e00);
346 
347  /* Stability check */
348  for (j = 1; j < LPC_ORDER; j++) {
349  temp = min_dist + cur_lsp[j - 1] - cur_lsp[j];
350  if (temp > 0) {
351  temp >>= 1;
352  cur_lsp[j - 1] -= temp;
353  cur_lsp[j] += temp;
354  }
355  }
356  stable = 1;
357  for (j = 1; j < LPC_ORDER; j++) {
358  temp = cur_lsp[j - 1] + min_dist - cur_lsp[j] - 4;
359  if (temp > 0) {
360  stable = 0;
361  break;
362  }
363  }
364  if (stable)
365  break;
366  }
367  if (!stable)
368  memcpy(cur_lsp, prev_lsp, LPC_ORDER * sizeof(*cur_lsp));
369 }
370 
377 #define MULL2(a, b) \
378  ((((a) >> 16) * (b) << 1) + (((a) & 0xffff) * (b) >> 15))
379 
385 static void lsp2lpc(int16_t *lpc)
386 {
387  int f1[LPC_ORDER / 2 + 1];
388  int f2[LPC_ORDER / 2 + 1];
389  int i, j;
390 
391  /* Calculate negative cosine */
392  for (j = 0; j < LPC_ORDER; j++) {
393  int index = lpc[j] >> 7;
394  int offset = lpc[j] & 0x7f;
395  int temp1 = cos_tab[index] << 16;
396  int temp2 = (cos_tab[index + 1] - cos_tab[index]) *
397  ((offset << 8) + 0x80) << 1;
398 
399  lpc[j] = -(av_sat_dadd32(1 << 15, temp1 + temp2) >> 16);
400  }
401 
402  /*
403  * Compute sum and difference polynomial coefficients
404  * (bitexact alternative to lsp2poly() in lsp.c)
405  */
406  /* Initialize with values in Q28 */
407  f1[0] = 1 << 28;
408  f1[1] = (lpc[0] << 14) + (lpc[2] << 14);
409  f1[2] = lpc[0] * lpc[2] + (2 << 28);
410 
411  f2[0] = 1 << 28;
412  f2[1] = (lpc[1] << 14) + (lpc[3] << 14);
413  f2[2] = lpc[1] * lpc[3] + (2 << 28);
414 
415  /*
416  * Calculate and scale the coefficients by 1/2 in
417  * each iteration for a final scaling factor of Q25
418  */
419  for (i = 2; i < LPC_ORDER / 2; i++) {
420  f1[i + 1] = f1[i - 1] + MULL2(f1[i], lpc[2 * i]);
421  f2[i + 1] = f2[i - 1] + MULL2(f2[i], lpc[2 * i + 1]);
422 
423  for (j = i; j >= 2; j--) {
424  f1[j] = MULL2(f1[j - 1], lpc[2 * i]) +
425  (f1[j] >> 1) + (f1[j - 2] >> 1);
426  f2[j] = MULL2(f2[j - 1], lpc[2 * i + 1]) +
427  (f2[j] >> 1) + (f2[j - 2] >> 1);
428  }
429 
430  f1[0] >>= 1;
431  f2[0] >>= 1;
432  f1[1] = ((lpc[2 * i] << 16 >> i) + f1[1]) >> 1;
433  f2[1] = ((lpc[2 * i + 1] << 16 >> i) + f2[1]) >> 1;
434  }
435 
436  /* Convert polynomial coefficients to LPC coefficients */
437  for (i = 0; i < LPC_ORDER / 2; i++) {
438  int64_t ff1 = f1[i + 1] + f1[i];
439  int64_t ff2 = f2[i + 1] - f2[i];
440 
441  lpc[i] = av_clipl_int32(((ff1 + ff2) << 3) + (1 << 15)) >> 16;
442  lpc[LPC_ORDER - i - 1] = av_clipl_int32(((ff1 - ff2) << 3) +
443  (1 << 15)) >> 16;
444  }
445 }
446 
455 static void lsp_interpolate(int16_t *lpc, int16_t *cur_lsp, int16_t *prev_lsp)
456 {
457  int i;
458  int16_t *lpc_ptr = lpc;
459 
460  /* cur_lsp * 0.25 + prev_lsp * 0.75 */
461  ff_acelp_weighted_vector_sum(lpc, cur_lsp, prev_lsp,
462  4096, 12288, 1 << 13, 14, LPC_ORDER);
463  ff_acelp_weighted_vector_sum(lpc + LPC_ORDER, cur_lsp, prev_lsp,
464  8192, 8192, 1 << 13, 14, LPC_ORDER);
465  ff_acelp_weighted_vector_sum(lpc + 2 * LPC_ORDER, cur_lsp, prev_lsp,
466  12288, 4096, 1 << 13, 14, LPC_ORDER);
467  memcpy(lpc + 3 * LPC_ORDER, cur_lsp, LPC_ORDER * sizeof(*lpc));
468 
469  for (i = 0; i < SUBFRAMES; i++) {
470  lsp2lpc(lpc_ptr);
471  lpc_ptr += LPC_ORDER;
472  }
473 }
474 
478 static void gen_dirac_train(int16_t *buf, int pitch_lag)
479 {
480  int16_t vector[SUBFRAME_LEN];
481  int i, j;
482 
483  memcpy(vector, buf, SUBFRAME_LEN * sizeof(*vector));
484  for (i = pitch_lag; i < SUBFRAME_LEN; i += pitch_lag) {
485  for (j = 0; j < SUBFRAME_LEN - i; j++)
486  buf[i + j] += vector[j];
487  }
488 }
489 
499 static void gen_fcb_excitation(int16_t *vector, G723_1_Subframe *subfrm,
500  enum Rate cur_rate, int pitch_lag, int index)
501 {
502  int temp, i, j;
503 
504  memset(vector, 0, SUBFRAME_LEN * sizeof(*vector));
505 
506  if (cur_rate == RATE_6300) {
507  if (subfrm->pulse_pos >= max_pos[index])
508  return;
509 
510  /* Decode amplitudes and positions */
511  j = PULSE_MAX - pulses[index];
512  temp = subfrm->pulse_pos;
513  for (i = 0; i < SUBFRAME_LEN / GRID_SIZE; i++) {
514  temp -= combinatorial_table[j][i];
515  if (temp >= 0)
516  continue;
517  temp += combinatorial_table[j++][i];
518  if (subfrm->pulse_sign & (1 << (PULSE_MAX - j))) {
519  vector[subfrm->grid_index + GRID_SIZE * i] =
520  -fixed_cb_gain[subfrm->amp_index];
521  } else {
522  vector[subfrm->grid_index + GRID_SIZE * i] =
523  fixed_cb_gain[subfrm->amp_index];
524  }
525  if (j == PULSE_MAX)
526  break;
527  }
528  if (subfrm->dirac_train == 1)
529  gen_dirac_train(vector, pitch_lag);
530  } else { /* 5300 bps */
531  int cb_gain = fixed_cb_gain[subfrm->amp_index];
532  int cb_shift = subfrm->grid_index;
533  int cb_sign = subfrm->pulse_sign;
534  int cb_pos = subfrm->pulse_pos;
535  int offset, beta, lag;
536 
537  for (i = 0; i < 8; i += 2) {
538  offset = ((cb_pos & 7) << 3) + cb_shift + i;
539  vector[offset] = (cb_sign & 1) ? cb_gain : -cb_gain;
540  cb_pos >>= 3;
541  cb_sign >>= 1;
542  }
543 
544  /* Enhance harmonic components */
545  lag = pitch_contrib[subfrm->ad_cb_gain << 1] + pitch_lag +
546  subfrm->ad_cb_lag - 1;
547  beta = pitch_contrib[(subfrm->ad_cb_gain << 1) + 1];
548 
549  if (lag < SUBFRAME_LEN - 2) {
550  for (i = lag; i < SUBFRAME_LEN; i++)
551  vector[i] += beta * vector[i - lag] >> 15;
552  }
553  }
554 }
555 
559 static void get_residual(int16_t *residual, int16_t *prev_excitation, int lag)
560 {
561  int offset = PITCH_MAX - PITCH_ORDER / 2 - lag;
562  int i;
563 
564  residual[0] = prev_excitation[offset];
565  residual[1] = prev_excitation[offset + 1];
566 
567  offset += 2;
568  for (i = 2; i < SUBFRAME_LEN + PITCH_ORDER - 1; i++)
569  residual[i] = prev_excitation[offset + (i - 2) % lag];
570 }
571 
572 static int dot_product(const int16_t *a, const int16_t *b, int length)
573 {
574  int i, sum = 0;
575 
576  for (i = 0; i < length; i++) {
577  int prod = a[i] * b[i];
578  sum = av_sat_dadd32(sum, prod);
579  }
580  return sum;
581 }
582 
586 static void gen_acb_excitation(int16_t *vector, int16_t *prev_excitation,
587  int pitch_lag, G723_1_Subframe *subfrm,
588  enum Rate cur_rate)
589 {
590  int16_t residual[SUBFRAME_LEN + PITCH_ORDER - 1];
591  const int16_t *cb_ptr;
592  int lag = pitch_lag + subfrm->ad_cb_lag - 1;
593 
594  int i;
595  int sum;
596 
597  get_residual(residual, prev_excitation, lag);
598 
599  /* Select quantization table */
600  if (cur_rate == RATE_6300 && pitch_lag < SUBFRAME_LEN - 2)
601  cb_ptr = adaptive_cb_gain85;
602  else
603  cb_ptr = adaptive_cb_gain170;
604 
605  /* Calculate adaptive vector */
606  cb_ptr += subfrm->ad_cb_gain * 20;
607  for (i = 0; i < SUBFRAME_LEN; i++) {
608  sum = dot_product(residual + i, cb_ptr, PITCH_ORDER);
609  vector[i] = av_sat_dadd32(1 << 15, sum) >> 16;
610  }
611 }
612 
623 static int autocorr_max(const int16_t *buf, int offset, int *ccr_max,
624  int pitch_lag, int length, int dir)
625 {
626  int limit, ccr, lag = 0;
627  int i;
628 
629  pitch_lag = FFMIN(PITCH_MAX - 3, pitch_lag);
630  if (dir > 0)
631  limit = FFMIN(FRAME_LEN + PITCH_MAX - offset - length, pitch_lag + 3);
632  else
633  limit = pitch_lag + 3;
634 
635  for (i = pitch_lag - 3; i <= limit; i++) {
636  ccr = dot_product(buf, buf + dir * i, length);
637 
638  if (ccr > *ccr_max) {
639  *ccr_max = ccr;
640  lag = i;
641  }
642  }
643  return lag;
644 }
645 
656 static void comp_ppf_gains(int lag, PPFParam *ppf, enum Rate cur_rate,
657  int tgt_eng, int ccr, int res_eng)
658 {
659  int pf_residual; /* square of postfiltered residual */
660  int temp1, temp2;
661 
662  ppf->index = lag;
663 
664  temp1 = tgt_eng * res_eng >> 1;
665  temp2 = ccr * ccr << 1;
666 
667  if (temp2 > temp1) {
668  if (ccr >= res_eng) {
669  ppf->opt_gain = ppf_gain_weight[cur_rate];
670  } else {
671  ppf->opt_gain = (ccr << 15) / res_eng *
672  ppf_gain_weight[cur_rate] >> 15;
673  }
674  /* pf_res^2 = tgt_eng + 2*ccr*gain + res_eng*gain^2 */
675  temp1 = (tgt_eng << 15) + (ccr * ppf->opt_gain << 1);
676  temp2 = (ppf->opt_gain * ppf->opt_gain >> 15) * res_eng;
677  pf_residual = av_sat_add32(temp1, temp2 + (1 << 15)) >> 16;
678 
679  if (tgt_eng >= pf_residual << 1) {
680  temp1 = 0x7fff;
681  } else {
682  temp1 = (tgt_eng << 14) / pf_residual;
683  }
684 
685  /* scaling_gain = sqrt(tgt_eng/pf_res^2) */
686  ppf->sc_gain = square_root(temp1 << 16);
687  } else {
688  ppf->opt_gain = 0;
689  ppf->sc_gain = 0x7fff;
690  }
691 
692  ppf->opt_gain = av_clip_int16(ppf->opt_gain * ppf->sc_gain >> 15);
693 }
694 
704 static void comp_ppf_coeff(G723_1_Context *p, int offset, int pitch_lag,
705  PPFParam *ppf, enum Rate cur_rate)
706 {
707 
708  int16_t scale;
709  int i;
710  int temp1, temp2;
711 
712  /*
713  * 0 - target energy
714  * 1 - forward cross-correlation
715  * 2 - forward residual energy
716  * 3 - backward cross-correlation
717  * 4 - backward residual energy
718  */
719  int energy[5] = {0, 0, 0, 0, 0};
720  int16_t *buf = p->audio + LPC_ORDER + offset;
721  int fwd_lag = autocorr_max(buf, offset, &energy[1], pitch_lag,
722  SUBFRAME_LEN, 1);
723  int back_lag = autocorr_max(buf, offset, &energy[3], pitch_lag,
724  SUBFRAME_LEN, -1);
725 
726  ppf->index = 0;
727  ppf->opt_gain = 0;
728  ppf->sc_gain = 0x7fff;
729 
730  /* Case 0, Section 3.6 */
731  if (!back_lag && !fwd_lag)
732  return;
733 
734  /* Compute target energy */
735  energy[0] = dot_product(buf, buf, SUBFRAME_LEN);
736 
737  /* Compute forward residual energy */
738  if (fwd_lag)
739  energy[2] = dot_product(buf + fwd_lag, buf + fwd_lag, SUBFRAME_LEN);
740 
741  /* Compute backward residual energy */
742  if (back_lag)
743  energy[4] = dot_product(buf - back_lag, buf - back_lag, SUBFRAME_LEN);
744 
745  /* Normalize and shorten */
746  temp1 = 0;
747  for (i = 0; i < 5; i++)
748  temp1 = FFMAX(energy[i], temp1);
749 
750  scale = normalize_bits(temp1, 31);
751  for (i = 0; i < 5; i++)
752  energy[i] = (energy[i] << scale) >> 16;
753 
754  if (fwd_lag && !back_lag) { /* Case 1 */
755  comp_ppf_gains(fwd_lag, ppf, cur_rate, energy[0], energy[1],
756  energy[2]);
757  } else if (!fwd_lag) { /* Case 2 */
758  comp_ppf_gains(-back_lag, ppf, cur_rate, energy[0], energy[3],
759  energy[4]);
760  } else { /* Case 3 */
761 
762  /*
763  * Select the largest of energy[1]^2/energy[2]
764  * and energy[3]^2/energy[4]
765  */
766  temp1 = energy[4] * ((energy[1] * energy[1] + (1 << 14)) >> 15);
767  temp2 = energy[2] * ((energy[3] * energy[3] + (1 << 14)) >> 15);
768  if (temp1 >= temp2) {
769  comp_ppf_gains(fwd_lag, ppf, cur_rate, energy[0], energy[1],
770  energy[2]);
771  } else {
772  comp_ppf_gains(-back_lag, ppf, cur_rate, energy[0], energy[3],
773  energy[4]);
774  }
775  }
776 }
777 
788 static int comp_interp_index(G723_1_Context *p, int pitch_lag,
789  int *exc_eng, int *scale)
790 {
791  int offset = PITCH_MAX + 2 * SUBFRAME_LEN;
792  int16_t *buf = p->audio + LPC_ORDER;
793 
794  int index, ccr, tgt_eng, best_eng, temp;
795 
796  *scale = scale_vector(buf, p->excitation, FRAME_LEN + PITCH_MAX);
797  buf += offset;
798 
799  /* Compute maximum backward cross-correlation */
800  ccr = 0;
801  index = autocorr_max(buf, offset, &ccr, pitch_lag, SUBFRAME_LEN * 2, -1);
802  ccr = av_sat_add32(ccr, 1 << 15) >> 16;
803 
804  /* Compute target energy */
805  tgt_eng = dot_product(buf, buf, SUBFRAME_LEN * 2);
806  *exc_eng = av_sat_add32(tgt_eng, 1 << 15) >> 16;
807 
808  if (ccr <= 0)
809  return 0;
810 
811  /* Compute best energy */
812  best_eng = dot_product(buf - index, buf - index, SUBFRAME_LEN * 2);
813  best_eng = av_sat_add32(best_eng, 1 << 15) >> 16;
814 
815  temp = best_eng * *exc_eng >> 3;
816 
817  if (temp < ccr * ccr)
818  return index;
819  else
820  return 0;
821 }
822 
832 static void residual_interp(int16_t *buf, int16_t *out, int lag,
833  int gain, int *rseed)
834 {
835  int i;
836  if (lag) { /* Voiced */
837  int16_t *vector_ptr = buf + PITCH_MAX;
838  /* Attenuate */
839  for (i = 0; i < lag; i++)
840  out[i] = vector_ptr[i - lag] * 3 >> 2;
841  av_memcpy_backptr((uint8_t*)(out + lag), lag * sizeof(*out),
842  (FRAME_LEN - lag) * sizeof(*out));
843  } else { /* Unvoiced */
844  for (i = 0; i < FRAME_LEN; i++) {
845  *rseed = *rseed * 521 + 259;
846  out[i] = gain * *rseed >> 15;
847  }
848  memset(buf, 0, (FRAME_LEN + PITCH_MAX) * sizeof(*buf));
849  }
850 }
851 
860 static inline void iir_filter(int16_t *fir_coef, int16_t *iir_coef,
861  int16_t *src, int *dest)
862 {
863  int m, n;
864 
865  for (m = 0; m < SUBFRAME_LEN; m++) {
866  int64_t filter = 0;
867  for (n = 1; n <= LPC_ORDER; n++) {
868  filter -= fir_coef[n - 1] * src[m - n] -
869  iir_coef[n - 1] * (dest[m - n] >> 16);
870  }
871 
872  dest[m] = av_clipl_int32((src[m] << 16) + (filter << 3) + (1 << 15));
873  }
874 }
875 
883 static void gain_scale(G723_1_Context *p, int16_t * buf, int energy)
884 {
885  int num, denom, gain, bits1, bits2;
886  int i;
887 
888  num = energy;
889  denom = 0;
890  for (i = 0; i < SUBFRAME_LEN; i++) {
891  int temp = buf[i] >> 2;
892  temp *= temp;
893  denom = av_sat_dadd32(denom, temp);
894  }
895 
896  if (num && denom) {
897  bits1 = normalize_bits(num, 31);
898  bits2 = normalize_bits(denom, 31);
899  num = num << bits1 >> 1;
900  denom <<= bits2;
901 
902  bits2 = 5 + bits1 - bits2;
903  bits2 = FFMAX(0, bits2);
904 
905  gain = (num >> 1) / (denom >> 16);
906  gain = square_root(gain << 16 >> bits2);
907  } else {
908  gain = 1 << 12;
909  }
910 
911  for (i = 0; i < SUBFRAME_LEN; i++) {
912  p->pf_gain = (15 * p->pf_gain + gain + (1 << 3)) >> 4;
913  buf[i] = av_clip_int16((buf[i] * (p->pf_gain + (p->pf_gain >> 4)) +
914  (1 << 10)) >> 11);
915  }
916 }
917 
926 static void formant_postfilter(G723_1_Context *p, int16_t *lpc,
927  int16_t *buf, int16_t *dst)
928 {
929  int16_t filter_coef[2][LPC_ORDER];
930  int filter_signal[LPC_ORDER + FRAME_LEN], *signal_ptr;
931  int i, j, k;
932 
933  memcpy(buf, p->fir_mem, LPC_ORDER * sizeof(*buf));
934  memcpy(filter_signal, p->iir_mem, LPC_ORDER * sizeof(*filter_signal));
935 
936  for (i = LPC_ORDER, j = 0; j < SUBFRAMES; i += SUBFRAME_LEN, j++) {
937  for (k = 0; k < LPC_ORDER; k++) {
938  filter_coef[0][k] = (-lpc[k] * postfilter_tbl[0][k] +
939  (1 << 14)) >> 15;
940  filter_coef[1][k] = (-lpc[k] * postfilter_tbl[1][k] +
941  (1 << 14)) >> 15;
942  }
943  iir_filter(filter_coef[0], filter_coef[1], buf + i,
944  filter_signal + i);
945  lpc += LPC_ORDER;
946  }
947 
948  memcpy(p->fir_mem, buf + FRAME_LEN, LPC_ORDER * sizeof(*p->fir_mem));
949  memcpy(p->iir_mem, filter_signal + FRAME_LEN,
950  LPC_ORDER * sizeof(*p->iir_mem));
951 
952  buf += LPC_ORDER;
953  signal_ptr = filter_signal + LPC_ORDER;
954  for (i = 0; i < SUBFRAMES; i++) {
955  int temp;
956  int auto_corr[2];
957  int scale, energy;
958 
959  /* Normalize */
960  scale = scale_vector(dst, buf, SUBFRAME_LEN);
961 
962  /* Compute auto correlation coefficients */
963  auto_corr[0] = dot_product(dst, dst + 1, SUBFRAME_LEN - 1);
964  auto_corr[1] = dot_product(dst, dst, SUBFRAME_LEN);
965 
966  /* Compute reflection coefficient */
967  temp = auto_corr[1] >> 16;
968  if (temp) {
969  temp = (auto_corr[0] >> 2) / temp;
970  }
971  p->reflection_coef = (3 * p->reflection_coef + temp + 2) >> 2;
972  temp = -p->reflection_coef >> 1 & ~3;
973 
974  /* Compensation filter */
975  for (j = 0; j < SUBFRAME_LEN; j++) {
976  dst[j] = av_sat_dadd32(signal_ptr[j],
977  (signal_ptr[j - 1] >> 16) * temp) >> 16;
978  }
979 
980  /* Compute normalized signal energy */
981  temp = 2 * scale + 4;
982  if (temp < 0) {
983  energy = av_clipl_int32((int64_t)auto_corr[1] << -temp);
984  } else
985  energy = auto_corr[1] >> temp;
986 
987  gain_scale(p, dst, energy);
988 
989  buf += SUBFRAME_LEN;
990  signal_ptr += SUBFRAME_LEN;
991  dst += SUBFRAME_LEN;
992  }
993 }
994 
995 static int sid_gain_to_lsp_index(int gain)
996 {
997  if (gain < 0x10)
998  return gain << 6;
999  else if (gain < 0x20)
1000  return gain - 8 << 7;
1001  else
1002  return gain - 20 << 8;
1003 }
1004 
1005 static inline int cng_rand(int *state, int base)
1006 {
1007  *state = (*state * 521 + 259) & 0xFFFF;
1008  return (*state & 0x7FFF) * base >> 15;
1009 }
1010 
1012 {
1013  int i, shift, seg, seg2, t, val, val_add, x, y;
1014 
1015  shift = 16 - p->cur_gain * 2;
1016  if (shift > 0)
1017  t = p->sid_gain << shift;
1018  else
1019  t = p->sid_gain >> -shift;
1020  x = t * cng_filt[0] >> 16;
1021 
1022  if (x >= cng_bseg[2])
1023  return 0x3F;
1024 
1025  if (x >= cng_bseg[1]) {
1026  shift = 4;
1027  seg = 3;
1028  } else {
1029  shift = 3;
1030  seg = (x >= cng_bseg[0]);
1031  }
1032  seg2 = FFMIN(seg, 3);
1033 
1034  val = 1 << shift;
1035  val_add = val >> 1;
1036  for (i = 0; i < shift; i++) {
1037  t = seg * 32 + (val << seg2);
1038  t *= t;
1039  if (x >= t)
1040  val += val_add;
1041  else
1042  val -= val_add;
1043  val_add >>= 1;
1044  }
1045 
1046  t = seg * 32 + (val << seg2);
1047  y = t * t - x;
1048  if (y <= 0) {
1049  t = seg * 32 + (val + 1 << seg2);
1050  t = t * t - x;
1051  val = (seg2 - 1 << 4) + val;
1052  if (t >= y)
1053  val++;
1054  } else {
1055  t = seg * 32 + (val - 1 << seg2);
1056  t = t * t - x;
1057  val = (seg2 - 1 << 4) + val;
1058  if (t >= y)
1059  val--;
1060  }
1061 
1062  return val;
1063 }
1064 
1066 {
1067  int i, j, idx, t;
1068  int off[SUBFRAMES];
1069  int signs[SUBFRAMES / 2 * 11], pos[SUBFRAMES / 2 * 11];
1070  int tmp[SUBFRAME_LEN * 2];
1071  int16_t *vector_ptr;
1072  int64_t sum;
1073  int b0, c, delta, x, shift;
1074 
1075  p->pitch_lag[0] = cng_rand(&p->cng_random_seed, 21) + 123;
1076  p->pitch_lag[1] = cng_rand(&p->cng_random_seed, 19) + 123;
1077 
1078  for (i = 0; i < SUBFRAMES; i++) {
1079  p->subframe[i].ad_cb_gain = cng_rand(&p->cng_random_seed, 50) + 1;
1081  }
1082 
1083  for (i = 0; i < SUBFRAMES / 2; i++) {
1084  t = cng_rand(&p->cng_random_seed, 1 << 13);
1085  off[i * 2] = t & 1;
1086  off[i * 2 + 1] = ((t >> 1) & 1) + SUBFRAME_LEN;
1087  t >>= 2;
1088  for (j = 0; j < 11; j++) {
1089  signs[i * 11 + j] = (t & 1) * 2 - 1 << 14;
1090  t >>= 1;
1091  }
1092  }
1093 
1094  idx = 0;
1095  for (i = 0; i < SUBFRAMES; i++) {
1096  for (j = 0; j < SUBFRAME_LEN / 2; j++)
1097  tmp[j] = j;
1098  t = SUBFRAME_LEN / 2;
1099  for (j = 0; j < pulses[i]; j++, idx++) {
1100  int idx2 = cng_rand(&p->cng_random_seed, t);
1101 
1102  pos[idx] = tmp[idx2] * 2 + off[i];
1103  tmp[idx2] = tmp[--t];
1104  }
1105  }
1106 
1107  vector_ptr = p->audio + LPC_ORDER;
1108  memcpy(vector_ptr, p->prev_excitation,
1109  PITCH_MAX * sizeof(*p->excitation));
1110  for (i = 0; i < SUBFRAMES; i += 2) {
1111  gen_acb_excitation(vector_ptr, vector_ptr,
1112  p->pitch_lag[i >> 1], &p->subframe[i],
1113  p->cur_rate);
1114  gen_acb_excitation(vector_ptr + SUBFRAME_LEN,
1115  vector_ptr + SUBFRAME_LEN,
1116  p->pitch_lag[i >> 1], &p->subframe[i + 1],
1117  p->cur_rate);
1118 
1119  t = 0;
1120  for (j = 0; j < SUBFRAME_LEN * 2; j++)
1121  t |= FFABS(vector_ptr[j]);
1122  t = FFMIN(t, 0x7FFF);
1123  if (!t) {
1124  shift = 0;
1125  } else {
1126  shift = -10 + av_log2(t);
1127  if (shift < -2)
1128  shift = -2;
1129  }
1130  sum = 0;
1131  if (shift < 0) {
1132  for (j = 0; j < SUBFRAME_LEN * 2; j++) {
1133  t = vector_ptr[j] << -shift;
1134  sum += t * t;
1135  tmp[j] = t;
1136  }
1137  } else {
1138  for (j = 0; j < SUBFRAME_LEN * 2; j++) {
1139  t = vector_ptr[j] >> shift;
1140  sum += t * t;
1141  tmp[j] = t;
1142  }
1143  }
1144 
1145  b0 = 0;
1146  for (j = 0; j < 11; j++)
1147  b0 += tmp[pos[(i / 2) * 11 + j]] * signs[(i / 2) * 11 + j];
1148  b0 = b0 * 2 * 2979LL + (1 << 29) >> 30; // approximated division by 11
1149 
1150  c = p->cur_gain * (p->cur_gain * SUBFRAME_LEN >> 5);
1151  if (shift * 2 + 3 >= 0)
1152  c >>= shift * 2 + 3;
1153  else
1154  c <<= -(shift * 2 + 3);
1155  c = (av_clipl_int32(sum << 1) - c) * 2979LL >> 15;
1156 
1157  delta = b0 * b0 * 2 - c;
1158  if (delta <= 0) {
1159  x = -b0;
1160  } else {
1161  delta = square_root(delta);
1162  x = delta - b0;
1163  t = delta + b0;
1164  if (FFABS(t) < FFABS(x))
1165  x = -t;
1166  }
1167  shift++;
1168  if (shift < 0)
1169  x >>= -shift;
1170  else
1171  x <<= shift;
1172  x = av_clip(x, -10000, 10000);
1173 
1174  for (j = 0; j < 11; j++) {
1175  idx = (i / 2) * 11 + j;
1176  vector_ptr[pos[idx]] = av_clip_int16(vector_ptr[pos[idx]] +
1177  (x * signs[idx] >> 15));
1178  }
1179 
1180  /* copy decoded data to serve as a history for the next decoded subframes */
1181  memcpy(vector_ptr + PITCH_MAX, vector_ptr,
1182  sizeof(*vector_ptr) * SUBFRAME_LEN * 2);
1183  vector_ptr += SUBFRAME_LEN * 2;
1184  }
1185  /* Save the excitation for the next frame */
1186  memcpy(p->prev_excitation, p->audio + LPC_ORDER + FRAME_LEN,
1187  PITCH_MAX * sizeof(*p->excitation));
1188 }
1189 
1190 static int g723_1_decode_frame(AVCodecContext *avctx, void *data,
1191  int *got_frame_ptr, AVPacket *avpkt)
1192 {
1193  G723_1_Context *p = avctx->priv_data;
1194  const uint8_t *buf = avpkt->data;
1195  int buf_size = avpkt->size;
1196  int dec_mode = buf[0] & 3;
1197 
1198  PPFParam ppf[SUBFRAMES];
1199  int16_t cur_lsp[LPC_ORDER];
1200  int16_t lpc[SUBFRAMES * LPC_ORDER];
1201  int16_t acb_vector[SUBFRAME_LEN];
1202  int16_t *out;
1203  int bad_frame = 0, i, j, ret;
1204  int16_t *audio = p->audio;
1205 
1206  if (buf_size < frame_size[dec_mode]) {
1207  if (buf_size)
1208  av_log(avctx, AV_LOG_WARNING,
1209  "Expected %d bytes, got %d - skipping packet\n",
1210  frame_size[dec_mode], buf_size);
1211  *got_frame_ptr = 0;
1212  return buf_size;
1213  }
1214 
1215  if (unpack_bitstream(p, buf, buf_size) < 0) {
1216  bad_frame = 1;
1217  if (p->past_frame_type == ACTIVE_FRAME)
1219  else
1221  }
1222 
1223  p->frame.nb_samples = FRAME_LEN;
1224  if ((ret = ff_get_buffer(avctx, &p->frame)) < 0) {
1225  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1226  return ret;
1227  }
1228 
1229  out = (int16_t *)p->frame.data[0];
1230 
1231  if (p->cur_frame_type == ACTIVE_FRAME) {
1232  if (!bad_frame)
1233  p->erased_frames = 0;
1234  else if (p->erased_frames != 3)
1235  p->erased_frames++;
1236 
1237  inverse_quant(cur_lsp, p->prev_lsp, p->lsp_index, bad_frame);
1238  lsp_interpolate(lpc, cur_lsp, p->prev_lsp);
1239 
1240  /* Save the lsp_vector for the next frame */
1241  memcpy(p->prev_lsp, cur_lsp, LPC_ORDER * sizeof(*p->prev_lsp));
1242 
1243  /* Generate the excitation for the frame */
1244  memcpy(p->excitation, p->prev_excitation,
1245  PITCH_MAX * sizeof(*p->excitation));
1246  if (!p->erased_frames) {
1247  int16_t *vector_ptr = p->excitation + PITCH_MAX;
1248 
1249  /* Update interpolation gain memory */
1251  p->subframe[3].amp_index) >> 1];
1252  for (i = 0; i < SUBFRAMES; i++) {
1253  gen_fcb_excitation(vector_ptr, &p->subframe[i], p->cur_rate,
1254  p->pitch_lag[i >> 1], i);
1255  gen_acb_excitation(acb_vector, &p->excitation[SUBFRAME_LEN * i],
1256  p->pitch_lag[i >> 1], &p->subframe[i],
1257  p->cur_rate);
1258  /* Get the total excitation */
1259  for (j = 0; j < SUBFRAME_LEN; j++) {
1260  int v = av_clip_int16(vector_ptr[j] << 1);
1261  vector_ptr[j] = av_clip_int16(v + acb_vector[j]);
1262  }
1263  vector_ptr += SUBFRAME_LEN;
1264  }
1265 
1266  vector_ptr = p->excitation + PITCH_MAX;
1267 
1269  &p->sid_gain, &p->cur_gain);
1270 
1271  /* Peform pitch postfiltering */
1272  if (p->postfilter) {
1273  i = PITCH_MAX;
1274  for (j = 0; j < SUBFRAMES; i += SUBFRAME_LEN, j++)
1275  comp_ppf_coeff(p, i, p->pitch_lag[j >> 1],
1276  ppf + j, p->cur_rate);
1277 
1278  for (i = 0, j = 0; j < SUBFRAMES; i += SUBFRAME_LEN, j++)
1280  vector_ptr + i,
1281  vector_ptr + i + ppf[j].index,
1282  ppf[j].sc_gain,
1283  ppf[j].opt_gain,
1284  1 << 14, 15, SUBFRAME_LEN);
1285  } else {
1286  audio = vector_ptr - LPC_ORDER;
1287  }
1288 
1289  /* Save the excitation for the next frame */
1290  memcpy(p->prev_excitation, p->excitation + FRAME_LEN,
1291  PITCH_MAX * sizeof(*p->excitation));
1292  } else {
1293  p->interp_gain = (p->interp_gain * 3 + 2) >> 2;
1294  if (p->erased_frames == 3) {
1295  /* Mute output */
1296  memset(p->excitation, 0,
1297  (FRAME_LEN + PITCH_MAX) * sizeof(*p->excitation));
1298  memset(p->prev_excitation, 0,
1299  PITCH_MAX * sizeof(*p->excitation));
1300  memset(p->frame.data[0], 0,
1301  (FRAME_LEN + LPC_ORDER) * sizeof(int16_t));
1302  } else {
1303  int16_t *buf = p->audio + LPC_ORDER;
1304 
1305  /* Regenerate frame */
1307  p->interp_gain, &p->random_seed);
1308 
1309  /* Save the excitation for the next frame */
1310  memcpy(p->prev_excitation, buf + (FRAME_LEN - PITCH_MAX),
1311  PITCH_MAX * sizeof(*p->excitation));
1312  }
1313  }
1315  } else {
1316  if (p->cur_frame_type == SID_FRAME) {
1318  inverse_quant(p->sid_lsp, p->prev_lsp, p->lsp_index, 0);
1319  } else if (p->past_frame_type == ACTIVE_FRAME) {
1320  p->sid_gain = estimate_sid_gain(p);
1321  }
1322 
1323  if (p->past_frame_type == ACTIVE_FRAME)
1324  p->cur_gain = p->sid_gain;
1325  else
1326  p->cur_gain = (p->cur_gain * 7 + p->sid_gain) >> 3;
1327  generate_noise(p);
1328  lsp_interpolate(lpc, p->sid_lsp, p->prev_lsp);
1329  /* Save the lsp_vector for the next frame */
1330  memcpy(p->prev_lsp, p->sid_lsp, LPC_ORDER * sizeof(*p->prev_lsp));
1331  }
1332 
1334 
1335  memcpy(p->audio, p->synth_mem, LPC_ORDER * sizeof(*p->audio));
1336  for (i = LPC_ORDER, j = 0; j < SUBFRAMES; i += SUBFRAME_LEN, j++)
1337  ff_celp_lp_synthesis_filter(p->audio + i, &lpc[j * LPC_ORDER],
1338  audio + i, SUBFRAME_LEN, LPC_ORDER,
1339  0, 1, 1 << 12);
1340  memcpy(p->synth_mem, p->audio + FRAME_LEN, LPC_ORDER * sizeof(*p->audio));
1341 
1342  if (p->postfilter) {
1343  formant_postfilter(p, lpc, p->audio, out);
1344  } else { // if output is not postfiltered it should be scaled by 2
1345  for (i = 0; i < FRAME_LEN; i++)
1346  out[i] = av_clip_int16(p->audio[LPC_ORDER + i] << 1);
1347  }
1348 
1349  *got_frame_ptr = 1;
1350  *(AVFrame *)data = p->frame;
1351 
1352  return frame_size[dec_mode];
1353 }
1354 
1355 #define OFFSET(x) offsetof(G723_1_Context, x)
1356 #define AD AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1357 
1358 static const AVOption options[] = {
1359  { "postfilter", "postfilter on/off", OFFSET(postfilter), AV_OPT_TYPE_INT,
1360  { .i64 = 1 }, 0, 1, AD },
1361  { NULL }
1362 };
1363 
1364 
1365 static const AVClass g723_1dec_class = {
1366  .class_name = "G.723.1 decoder",
1367  .item_name = av_default_item_name,
1368  .option = options,
1369  .version = LIBAVUTIL_VERSION_INT,
1370 };
1371 
1373  .name = "g723_1",
1374  .type = AVMEDIA_TYPE_AUDIO,
1375  .id = AV_CODEC_ID_G723_1,
1376  .priv_data_size = sizeof(G723_1_Context),
1379  .long_name = NULL_IF_CONFIG_SMALL("G.723.1"),
1380  .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1381  .priv_class = &g723_1dec_class,
1382 };
static void lsp2lpc(int16_t *lpc)
Convert LSP frequencies to LPC coefficients.
Definition: g723_1.c:385
int postfilter
Definition: g723_1.c:105
int dirac_train
Definition: g723_1.c:61
This structure describes decoded (raw) audio or video data.
Definition: avcodec.h:989
static const int16_t lsp_band0[LSP_CB_SIZE][3]
Definition: g723_1_data.h:127
int ad_cb_gain
Definition: g723_1.c:60
static const int cng_bseg[3]
Definition: g723_1_data.h:1198
AVOption.
Definition: opt.h:233
FrameType
G723.1 frame types.
Definition: g723_1.c:44
static unsigned int get_bits(GetBitContext *s, int n)
Read 1-25 bits.
Definition: get_bits.h:237
static const int16_t lsp_band2[LSP_CB_SIZE][4]
Definition: g723_1_data.h:305
memory handling functions
AVFrame * coded_frame
the picture in the bitstream
Definition: avcodec.h:2725
G723.1 unpacked data subframe.
Definition: g723_1.c:58
int ff_celp_lp_synthesis_filter(int16_t *out, const int16_t *filter_coeffs, const int16_t *in, int buffer_length, int filter_length, int stop_on_overflow, int shift, int rounder)
LP synthesis filter.
Definition: celp_filters.c:59
static int scale_vector(int16_t *dst, const int16_t *vector, int length)
Scale vector contents based on the largest of their absolutes.
Definition: g723_1.c:283
static const int cng_filt[4]
Definition: g723_1_data.h:1196
uint8_t lsp_index[LSP_BANDS]
Definition: g723_1.c:85
static int normalize_bits(int num, int width)
Calculate the number of left-shifts required for normalizing the input.
Definition: g723_1.c:275
int size
Definition: avcodec.h:916
static int autocorr_max(const int16_t *buf, int offset, int *ccr_max, int pitch_lag, int length, int dir)
Estimate maximum auto-correlation around pitch lag.
Definition: g723_1.c:623
static void iir_filter(int16_t *fir_coef, int16_t *iir_coef, int16_t *src, int *dest)
Perform IIR filtering.
Definition: g723_1.c:860
#define PITCH_MAX
Definition: g723_1_data.h:40
static int comp_interp_index(G723_1_Context *p, int pitch_lag, int *exc_eng, int *scale)
Classify frames as voiced/unvoiced.
Definition: g723_1.c:788
int interp_gain
Definition: g723_1.c:100
#define SUBFRAMES
Definition: g723_1_data.h:33
static const AVOption options[]
Definition: g723_1.c:1358
static void gen_dirac_train(int16_t *buf, int pitch_lag)
Generate a train of dirac functions with period as pitch lag.
Definition: g723_1.c:478
static const AVClass g723_1dec_class
Definition: g723_1.c:1365
static void comp_ppf_coeff(G723_1_Context *p, int offset, int pitch_lag, PPFParam *ppf, enum Rate cur_rate)
Calculate pitch postfilter parameters.
Definition: g723_1.c:704
signed 16 bits
Definition: samplefmt.h:52
static const uint8_t frame_size[4]
Definition: g723_1.c:31
AVCodec.
Definition: avcodec.h:2960
#define LSP_BANDS
Definition: g723_1_data.h:37
static void postfilter(AMRContext *p, float *lpc, float *buf_out)
Perform adaptive post-filtering to enhance the quality of the speech.
Definition: amrnbdec.c:890
#define GRID_SIZE
Definition: g723_1_data.h:42
int16_t prev_lsp[LPC_ORDER]
Definition: g723_1.c:89
const char * class_name
The name of the class; usually it is the same name as the context structure type to which the AVClass...
Definition: log.h:38
static int decode(MimicContext *ctx, int quality, int num_coeffs, int is_iframe)
Definition: mimic.c:228
static int g723_1_decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr, AVPacket *avpkt)
Definition: g723_1.c:1190
int16_t prev_excitation[PITCH_MAX]
Definition: g723_1.c:91
#define SUBFRAME_LEN
Definition: g723_1_data.h:34
uint8_t bits
Definition: crc.c:31
enum AVSampleFormat sample_fmt
audio sample format
Definition: avcodec.h:2112
uint8_t
float delta
AVOptions.
static const int16_t lsp_band1[LSP_CB_SIZE][3]
Definition: g723_1_data.h:216
static void residual_interp(int16_t *buf, int16_t *out, int lag, int gain, int *rseed)
Peform residual interpolation based on frame classification.
Definition: g723_1.c:832
#define b
Definition: input.c:52
int random_seed
Definition: g723_1.c:97
int pulse_sign
Definition: g723_1.c:62
const char data[16]
Definition: mxf.c:66
static void formant_postfilter(G723_1_Context *p, int16_t *lpc, int16_t *buf, int16_t *dst)
Perform formant filtering.
Definition: g723_1.c:926
uint8_t * data
Definition: avcodec.h:915
static const uint8_t bits2[81]
Definition: aactab.c:118
bitstream reader API header.
static float t
static int init(AVCodecParserContext *s)
Definition: h264_parser.c:335
#define CNG_RANDOM_SEED
Definition: g723_1.c:39
static void get_residual(int16_t *residual, int16_t *prev_excitation, int lag)
Get delayed contribution from the previous excitation vector.
Definition: g723_1.c:559
int amp_index
Definition: g723_1.c:64
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification. ...
Definition: internal.h:88
enum FrameType cur_frame_type
Definition: g723_1.c:82
static const int16_t postfilter_tbl[2][LPC_ORDER]
Definition: g723_1_data.h:1187
void av_log(void *avcl, int level, const char *fmt,...)
Definition: log.c:146
int grid_index
Definition: g723_1.c:63
const char * name
Name of the codec implementation.
Definition: avcodec.h:2967
AVFrame frame
Definition: g723_1.c:79
static void lsp_interpolate(int16_t *lpc, int16_t *cur_lsp, int16_t *prev_lsp)
Quantize LSP frequencies by interpolation and convert them to the corresponding LPC coefficients...
Definition: g723_1.c:455
G.723.1 compatible decoder data tables.
static av_cold int g723_1_decode_init(AVCodecContext *avctx)
Definition: g723_1.c:110
static void comp_ppf_gains(int lag, PPFParam *ppf, enum Rate cur_rate, int tgt_eng, int ccr, int res_eng)
Calculate pitch postfilter optimal and scaling gains.
Definition: g723_1.c:656
int off
Definition: dsputil_bfin.c:28
uint64_t channel_layout
Audio channel layout.
Definition: avcodec.h:2165
static int estimate_sid_gain(G723_1_Context *p)
Definition: g723_1.c:1011
static void filter(MpegAudioContext *s, int ch, const short *samples, int incr)
Definition: mpegaudioenc.c:318
static const int16_t fixed_cb_gain[GAIN_LEVELS]
Definition: g723_1_data.h:536
void ff_acelp_weighted_vector_sum(int16_t *out, const int16_t *in_a, const int16_t *in_b, int16_t weight_coeff_a, int16_t weight_coeff_b, int16_t rounder, int shift, int length)
weighted sum of two vectors with rounding.
audio channel layout utility functions
G723_1_Subframe subframe[4]
Definition: g723_1.c:81
struct g723_1_context G723_1_Context
int ff_get_buffer(AVCodecContext *avctx, AVFrame *frame)
Get a buffer for a frame.
Definition: utils.c:464
enum FrameType past_frame_type
Definition: g723_1.c:83
static const int16_t ppf_gain_weight[2]
Definition: g723_1_data.h:50
#define PITCH_MIN
Definition: g723_1_data.h:39
int index
postfilter backward/forward lag
Definition: g723_1.c:72
#define OFFSET(x)
Definition: g723_1.c:1355
LIBAVUTIL_VERSION_INT
Definition: eval.c:52
static const int16_t adaptive_cb_gain85[85 *20]
Definition: g723_1_data.h:542
int sid_gain
Definition: g723_1.c:101
static int16_t square_root(int val)
Bitexact implementation of sqrt(val/2).
Definition: g723_1.c:254
static const float pred[4]
Definition: siprdata.h:259
static const int16_t adaptive_cb_gain170[170 *20]
Definition: g723_1_data.h:758
int16_t opt_gain
optimal gain
Definition: g723_1.c:73
NULL
Definition: eval.c:52
static int width
Definition: utils.c:156
external API header
static const int16_t cos_tab[COS_TBL_SIZE]
Definition: g723_1_data.h:59
int cng_random_seed
Definition: g723_1.c:98
#define PITCH_ORDER
Definition: g723_1_data.h:41
int sample_rate
samples per second
Definition: avcodec.h:2104
int16_t sid_lsp[LPC_ORDER]
Definition: g723_1.c:90
av_default_item_name
Definition: dnxhdenc.c:43
#define AD
Definition: g723_1.c:1356
main external API structure.
Definition: avcodec.h:1339
#define FASTDIV(a, b)
Definition: mathops.h:195
#define PULSE_MAX
Definition: g723_1_data.h:43
static void gen_acb_excitation(int16_t *vector, int16_t *prev_excitation, int pitch_lag, G723_1_Subframe *subfrm, enum Rate cur_rate)
Generate adaptive codebook excitation.
Definition: g723_1.c:586
Active speech.
Definition: g723_1.c:45
#define MULL2(a, b)
Bitexact implementation of 2ab scaled by 1/2^16.
Definition: g723_1.c:377
void avcodec_get_frame_defaults(AVFrame *frame)
Set the fields of the given AVFrame to default values.
Definition: utils.c:604
Describe the class of an AVClass context structure.
Definition: log.h:33
int16_t sc_gain
scaling gain
Definition: g723_1.c:74
static void skip_bits(GetBitContext *s, int n)
Definition: get_bits.h:260
int index
Definition: gxfenc.c:72
static const int16_t pitch_contrib[340]
Definition: g723_1_data.h:484
static int init_get_bits(GetBitContext *s, const uint8_t *buffer, int bit_size)
Initialize GetBitContext.
Definition: get_bits.h:372
static void gen_fcb_excitation(int16_t *vector, G723_1_Subframe *subfrm, enum Rate cur_rate, int pitch_lag, int index)
Generate fixed codebook excitation vector.
Definition: g723_1.c:499
static int dot_product(const int16_t *a, const int16_t *b, int length)
Definition: g723_1.c:572
int erased_frames
Definition: g723_1.c:87
static int sid_gain_to_lsp_index(int gain)
Definition: g723_1.c:995
static const int cng_adaptive_cb_lag[4]
Definition: g723_1_data.h:1194
static uint32_t state
Definition: trasher.c:27
static const uint16_t scale[4]
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
Definition: avcodec.h:997
#define LPC_ORDER
Definition: g723_1_data.h:36
AVCodec ff_g723_1_decoder
Definition: g723_1.c:1372
common internal api header.
Pitch postfilter parameters.
Definition: g723_1.c:71
int reflection_coef
Definition: g723_1.c:103
static int unpack_bitstream(G723_1_Context *p, const uint8_t *buf, int buf_size)
Unpack the frame into parameters.
Definition: g723_1.c:139
#define FRAME_LEN
Definition: g723_1_data.h:35
int iir_mem[LPC_ORDER]
Definition: g723_1.c:95
static const int32_t max_pos[4]
Definition: g723_1_data.h:534
int cur_gain
Definition: g723_1.c:102
int16_t fir_mem[LPC_ORDER]
Definition: g723_1.c:94
static const int16_t dc_lsp[LPC_ORDER]
Definition: g723_1_data.h:53
static void gain_scale(G723_1_Context *p, int16_t *buf, int energy)
Adjust gain of postfiltered signal.
Definition: g723_1.c:883
static void inverse_quant(int16_t *cur_lsp, int16_t *prev_lsp, uint8_t *lsp_index, int bad_frame)
Perform inverse quantization of LSP frequencies.
Definition: g723_1.c:309
void * priv_data
Definition: avcodec.h:1382
int16_t excitation[PITCH_MAX+FRAME_LEN+4]
Definition: g723_1.c:92
static int cng_rand(int *state, int base)
Definition: g723_1.c:1005
int16_t audio[FRAME_LEN+LPC_ORDER+PITCH_MAX+4]
Definition: g723_1.c:107
int interp_index
Definition: g723_1.c:99
int channels
number of audio channels
Definition: avcodec.h:2105
#define av_log2
Definition: intmath.h:85
int pulse_pos
Definition: g723_1.c:65
int16_t synth_mem[LPC_ORDER]
Definition: g723_1.c:93
#define GAIN_LEVELS
Definition: g723_1_data.h:44
static void generate_noise(G723_1_Context *p)
Definition: g723_1.c:1065
static const int8_t pulses[4]
Definition: g723_1_data.h:531
void av_memcpy_backptr(uint8_t *dst, int back, int cnt)
deliberately overlapping memcpy implementation
Definition: mem.c:252
Silence Insertion Descriptor frame.
Definition: g723_1.c:46
Rate
Definition: g723_1.c:50
#define AV_CH_LAYOUT_MONO
This structure stores compressed data.
Definition: avcodec.h:898
int nb_samples
number of audio samples (per channel) described by this frame
Definition: avcodec.h:1042
enum Rate cur_rate
Definition: g723_1.c:84
static const uint8_t bits1[81]
Definition: aactab.c:95
int ad_cb_lag
adaptive codebook lag
Definition: g723_1.c:59
static const int32_t combinatorial_table[PULSE_MAX][SUBFRAME_LEN/GRID_SIZE]
Definition: g723_1_data.h:440
int pitch_lag[2]
Definition: g723_1.c:86
if(!(ptr_align%ac->ptr_align)&&samples_align >=aligned_len)