twinvq.c
Go to the documentation of this file.
1 /*
2  * TwinVQ decoder
3  * Copyright (c) 2009 Vitor Sessak
4  *
5  * This file is part of Libav.
6  *
7  * Libav is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * Libav is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #include "avcodec.h"
23 #include "internal.h"
24 #include "get_bits.h"
25 #include "dsputil.h"
26 #include "fft.h"
27 #include "lsp.h"
28 #include "sinewin.h"
29 
30 #include <math.h>
31 #include <stdint.h>
32 
33 #include "twinvq_data.h"
34 
35 enum FrameType {
36  FT_SHORT = 0,
40 };
41 
45 struct FrameMode {
46  uint8_t sub;
47  const uint16_t *bark_tab;
48 
50  uint8_t bark_env_size;
51 
52  const int16_t *bark_cb;
53  uint8_t bark_n_coef;
54  uint8_t bark_n_bit;
55 
57 
58  const int16_t *cb0;
59  const int16_t *cb1;
61 
62  uint8_t cb_len_read;
63 };
64 
69 typedef struct {
70  struct FrameMode fmode[3];
71 
72  uint16_t size;
73  uint8_t n_lsp;
74  const float *lspcodebook;
75 
76  /* number of bits of the different LSP CB coefficients */
77  uint8_t lsp_bit0;
78  uint8_t lsp_bit1;
79  uint8_t lsp_bit2;
80 
81  uint8_t lsp_split;
82  const int16_t *ppc_shape_cb;
83 
85  uint8_t ppc_period_bit;
86 
87  uint8_t ppc_shape_bit;
88  uint8_t ppc_shape_len;
89  uint8_t pgain_bit;
90 
92  uint16_t peak_per2wid;
93 } ModeTab;
94 
95 static const ModeTab mode_08_08 = {
96  {
97  { 8, bark_tab_s08_64, 10, tab.fcb08s , 1, 5, tab.cb0808s0, tab.cb0808s1, 18},
98  { 2, bark_tab_m08_256, 20, tab.fcb08m , 2, 5, tab.cb0808m0, tab.cb0808m1, 16},
99  { 1, bark_tab_l08_512, 30, tab.fcb08l , 3, 6, tab.cb0808l0, tab.cb0808l1, 17}
100  },
101  512 , 12, tab.lsp08, 1, 5, 3, 3, tab.shape08 , 8, 28, 20, 6, 40
102 };
103 
104 static const ModeTab mode_11_08 = {
105  {
106  { 8, bark_tab_s11_64, 10, tab.fcb11s , 1, 5, tab.cb1108s0, tab.cb1108s1, 29},
107  { 2, bark_tab_m11_256, 20, tab.fcb11m , 2, 5, tab.cb1108m0, tab.cb1108m1, 24},
108  { 1, bark_tab_l11_512, 30, tab.fcb11l , 3, 6, tab.cb1108l0, tab.cb1108l1, 27}
109  },
110  512 , 16, tab.lsp11, 1, 6, 4, 3, tab.shape11 , 9, 36, 30, 7, 90
111 };
112 
113 static const ModeTab mode_11_10 = {
114  {
115  { 8, bark_tab_s11_64, 10, tab.fcb11s , 1, 5, tab.cb1110s0, tab.cb1110s1, 21},
116  { 2, bark_tab_m11_256, 20, tab.fcb11m , 2, 5, tab.cb1110m0, tab.cb1110m1, 18},
117  { 1, bark_tab_l11_512, 30, tab.fcb11l , 3, 6, tab.cb1110l0, tab.cb1110l1, 20}
118  },
119  512 , 16, tab.lsp11, 1, 6, 4, 3, tab.shape11 , 9, 36, 30, 7, 90
120 };
121 
122 static const ModeTab mode_16_16 = {
123  {
124  { 8, bark_tab_s16_128, 10, tab.fcb16s , 1, 5, tab.cb1616s0, tab.cb1616s1, 16},
125  { 2, bark_tab_m16_512, 20, tab.fcb16m , 2, 5, tab.cb1616m0, tab.cb1616m1, 15},
126  { 1, bark_tab_l16_1024,30, tab.fcb16l , 3, 6, tab.cb1616l0, tab.cb1616l1, 16}
127  },
128  1024, 16, tab.lsp16, 1, 6, 4, 3, tab.shape16 , 9, 56, 60, 7, 180
129 };
130 
131 static const ModeTab mode_22_20 = {
132  {
133  { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2220s0, tab.cb2220s1, 18},
134  { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2220m0, tab.cb2220m1, 17},
135  { 1, bark_tab_l22_1024,32, tab.fcb22l_1, 4, 6, tab.cb2220l0, tab.cb2220l1, 18}
136  },
137  1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
138 };
139 
140 static const ModeTab mode_22_24 = {
141  {
142  { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2224s0, tab.cb2224s1, 15},
143  { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2224m0, tab.cb2224m1, 14},
144  { 1, bark_tab_l22_1024,32, tab.fcb22l_1, 4, 6, tab.cb2224l0, tab.cb2224l1, 15}
145  },
146  1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
147 };
148 
149 static const ModeTab mode_22_32 = {
150  {
151  { 4, bark_tab_s22_128, 10, tab.fcb22s_2, 1, 6, tab.cb2232s0, tab.cb2232s1, 11},
152  { 2, bark_tab_m22_256, 20, tab.fcb22m_2, 2, 6, tab.cb2232m0, tab.cb2232m1, 11},
153  { 1, bark_tab_l22_512, 32, tab.fcb22l_2, 4, 6, tab.cb2232l0, tab.cb2232l1, 12}
154  },
155  512 , 16, tab.lsp22_2, 1, 6, 4, 4, tab.shape22_2, 9, 56, 36, 7, 72
156 };
157 
158 static const ModeTab mode_44_40 = {
159  {
160  {16, bark_tab_s44_128, 10, tab.fcb44s , 1, 6, tab.cb4440s0, tab.cb4440s1, 18},
161  { 4, bark_tab_m44_512, 20, tab.fcb44m , 2, 6, tab.cb4440m0, tab.cb4440m1, 17},
162  { 1, bark_tab_l44_2048,40, tab.fcb44l , 4, 6, tab.cb4440l0, tab.cb4440l1, 17}
163  },
164  2048, 20, tab.lsp44, 1, 6, 4, 4, tab.shape44 , 9, 84, 54, 7, 432
165 };
166 
167 static const ModeTab mode_44_48 = {
168  {
169  {16, bark_tab_s44_128, 10, tab.fcb44s , 1, 6, tab.cb4448s0, tab.cb4448s1, 15},
170  { 4, bark_tab_m44_512, 20, tab.fcb44m , 2, 6, tab.cb4448m0, tab.cb4448m1, 14},
171  { 1, bark_tab_l44_2048,40, tab.fcb44l , 4, 6, tab.cb4448l0, tab.cb4448l1, 14}
172  },
173  2048, 20, tab.lsp44, 1, 6, 4, 4, tab.shape44 , 9, 84, 54, 7, 432
174 };
175 
176 typedef struct TwinContext {
181 
182  const ModeTab *mtab;
183 
184  // history
185  float lsp_hist[2][20];
186  float bark_hist[3][2][40];
187 
188  // bitstream parameters
189  int16_t permut[4][4096];
190  uint8_t length[4][2];
191  uint8_t length_change[4];
192  uint8_t bits_main_spec[2][4][2];
194  int n_div[4];
195 
196  float *spectrum;
197  float *curr_frame;
198  float *prev_frame;
201 
202  float *cos_tabs[3];
203 
204  // scratch buffers
205  float *tmp_buf;
206 } TwinContext;
207 
208 #define PPC_SHAPE_CB_SIZE 64
209 #define PPC_SHAPE_LEN_MAX 60
210 #define SUB_AMP_MAX 4500.0
211 #define MULAW_MU 100.0
212 #define GAIN_BITS 8
213 #define AMP_MAX 13000.0
214 #define SUB_GAIN_BITS 5
215 #define WINDOW_TYPE_BITS 4
216 #define PGAIN_MU 200
217 #define LSP_COEFS_MAX 20
218 #define LSP_SPLIT_MAX 4
219 #define CHANNELS_MAX 2
220 #define SUBBLOCKS_MAX 16
221 #define BARK_N_COEF_MAX 4
222 
224 static void memset_float(float *buf, float val, int size)
225 {
226  while (size--)
227  *buf++ = val;
228 }
229 
242 static float eval_lpc_spectrum(const float *lsp, float cos_val, int order)
243 {
244  int j;
245  float p = 0.5f;
246  float q = 0.5f;
247  float two_cos_w = 2.0f*cos_val;
248 
249  for (j = 0; j + 1 < order; j += 2*2) {
250  // Unroll the loop once since order is a multiple of four
251  q *= lsp[j ] - two_cos_w;
252  p *= lsp[j+1] - two_cos_w;
253 
254  q *= lsp[j+2] - two_cos_w;
255  p *= lsp[j+3] - two_cos_w;
256  }
257 
258  p *= p * (2.0f - two_cos_w);
259  q *= q * (2.0f + two_cos_w);
260 
261  return 0.5 / (p + q);
262 }
263 
267 static void eval_lpcenv(TwinContext *tctx, const float *cos_vals, float *lpc)
268 {
269  int i;
270  const ModeTab *mtab = tctx->mtab;
271  int size_s = mtab->size / mtab->fmode[FT_SHORT].sub;
272 
273  for (i = 0; i < size_s/2; i++) {
274  float cos_i = tctx->cos_tabs[0][i];
275  lpc[i] = eval_lpc_spectrum(cos_vals, cos_i, mtab->n_lsp);
276  lpc[size_s-i-1] = eval_lpc_spectrum(cos_vals, -cos_i, mtab->n_lsp);
277  }
278 }
279 
280 static void interpolate(float *out, float v1, float v2, int size)
281 {
282  int i;
283  float step = (v1 - v2)/(size + 1);
284 
285  for (i = 0; i < size; i++) {
286  v2 += step;
287  out[i] = v2;
288  }
289 }
290 
291 static inline float get_cos(int idx, int part, const float *cos_tab, int size)
292 {
293  return part ? -cos_tab[size - idx - 1] :
294  cos_tab[ idx ];
295 }
296 
311 static inline void eval_lpcenv_or_interp(TwinContext *tctx,
312  enum FrameType ftype,
313  float *out, const float *in,
314  int size, int step, int part)
315 {
316  int i;
317  const ModeTab *mtab = tctx->mtab;
318  const float *cos_tab = tctx->cos_tabs[ftype];
319 
320  // Fill the 's'
321  for (i = 0; i < size; i += step)
322  out[i] =
324  get_cos(i, part, cos_tab, size),
325  mtab->n_lsp);
326 
327  // Fill the 'iiiibiiii'
328  for (i = step; i <= size - 2*step; i += step) {
329  if (out[i + step] + out[i - step] > 1.95*out[i] ||
330  out[i + step] >= out[i - step]) {
331  interpolate(out + i - step + 1, out[i], out[i-step], step - 1);
332  } else {
333  out[i - step/2] =
335  get_cos(i-step/2, part, cos_tab, size),
336  mtab->n_lsp);
337  interpolate(out + i - step + 1, out[i-step/2], out[i-step ], step/2 - 1);
338  interpolate(out + i - step/2 + 1, out[i ], out[i-step/2], step/2 - 1);
339  }
340  }
341 
342  interpolate(out + size - 2*step + 1, out[size-step], out[size - 2*step], step - 1);
343 }
344 
345 static void eval_lpcenv_2parts(TwinContext *tctx, enum FrameType ftype,
346  const float *buf, float *lpc,
347  int size, int step)
348 {
349  eval_lpcenv_or_interp(tctx, ftype, lpc , buf, size/2, step, 0);
350  eval_lpcenv_or_interp(tctx, ftype, lpc + size/2, buf, size/2, 2*step, 1);
351 
352  interpolate(lpc+size/2-step+1, lpc[size/2], lpc[size/2-step], step);
353 
354  memset_float(lpc + size - 2*step + 1, lpc[size - 2*step], 2*step - 1);
355 }
356 
362 static void dequant(TwinContext *tctx, GetBitContext *gb, float *out,
363  enum FrameType ftype,
364  const int16_t *cb0, const int16_t *cb1, int cb_len)
365 {
366  int pos = 0;
367  int i, j;
368 
369  for (i = 0; i < tctx->n_div[ftype]; i++) {
370  int tmp0, tmp1;
371  int sign0 = 1;
372  int sign1 = 1;
373  const int16_t *tab0, *tab1;
374  int length = tctx->length[ftype][i >= tctx->length_change[ftype]];
375  int bitstream_second_part = (i >= tctx->bits_main_spec_change[ftype]);
376 
377  int bits = tctx->bits_main_spec[0][ftype][bitstream_second_part];
378  if (bits == 7) {
379  if (get_bits1(gb))
380  sign0 = -1;
381  bits = 6;
382  }
383  tmp0 = get_bits(gb, bits);
384 
385  bits = tctx->bits_main_spec[1][ftype][bitstream_second_part];
386 
387  if (bits == 7) {
388  if (get_bits1(gb))
389  sign1 = -1;
390 
391  bits = 6;
392  }
393  tmp1 = get_bits(gb, bits);
394 
395  tab0 = cb0 + tmp0*cb_len;
396  tab1 = cb1 + tmp1*cb_len;
397 
398  for (j = 0; j < length; j++)
399  out[tctx->permut[ftype][pos+j]] = sign0*tab0[j] + sign1*tab1[j];
400 
401  pos += length;
402  }
403 
404 }
405 
406 static inline float mulawinv(float y, float clip, float mu)
407 {
408  y = av_clipf(y/clip, -1, 1);
409  return clip * FFSIGN(y) * (exp(log(1+mu) * fabs(y)) - 1) / mu;
410 }
411 
432 static int very_broken_op(int a, int b)
433 {
434  int x = a*b + 200;
435  int size;
436  const uint8_t *rtab;
437 
438  if (x%400 || b%5)
439  return x/400;
440 
441  x /= 400;
442 
443  size = tabs[b/5].size;
444  rtab = tabs[b/5].tab;
445  return x - rtab[size*av_log2(2*(x - 1)/size)+(x - 1)%size];
446 }
447 
453 static void add_peak(int period, int width, const float *shape,
454  float ppc_gain, float *speech, int len)
455 {
456  int i, j;
457 
458  const float *shape_end = shape + len;
459  int center;
460 
461  // First peak centered around zero
462  for (i = 0; i < width/2; i++)
463  speech[i] += ppc_gain * *shape++;
464 
465  for (i = 1; i < ROUNDED_DIV(len,width) ; i++) {
466  center = very_broken_op(period, i);
467  for (j = -width/2; j < (width+1)/2; j++)
468  speech[j+center] += ppc_gain * *shape++;
469  }
470 
471  // For the last block, be careful not to go beyond the end of the buffer
472  center = very_broken_op(period, i);
473  for (j = -width/2; j < (width + 1)/2 && shape < shape_end; j++)
474  speech[j+center] += ppc_gain * *shape++;
475 }
476 
477 static void decode_ppc(TwinContext *tctx, int period_coef, const float *shape,
478  float ppc_gain, float *speech)
479 {
480  const ModeTab *mtab = tctx->mtab;
481  int isampf = tctx->avctx->sample_rate/1000;
482  int ibps = tctx->avctx->bit_rate/(1000 * tctx->avctx->channels);
483  int min_period = ROUNDED_DIV( 40*2*mtab->size, isampf);
484  int max_period = ROUNDED_DIV(6*40*2*mtab->size, isampf);
485  int period_range = max_period - min_period;
486 
487  // This is actually the period multiplied by 400. It is just linearly coded
488  // between its maximum and minimum value.
489  int period = min_period +
490  ROUNDED_DIV(period_coef*period_range, (1 << mtab->ppc_period_bit) - 1);
491  int width;
492 
493  if (isampf == 22 && ibps == 32) {
494  // For some unknown reason, NTT decided to code this case differently...
495  width = ROUNDED_DIV((period + 800)* mtab->peak_per2wid, 400*mtab->size);
496  } else
497  width = (period )* mtab->peak_per2wid/(400*mtab->size);
498 
499  add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len);
500 }
501 
502 static void dec_gain(TwinContext *tctx, GetBitContext *gb, enum FrameType ftype,
503  float *out)
504 {
505  const ModeTab *mtab = tctx->mtab;
506  int i, j;
507  int sub = mtab->fmode[ftype].sub;
508  float step = AMP_MAX / ((1 << GAIN_BITS) - 1);
509  float sub_step = SUB_AMP_MAX / ((1 << SUB_GAIN_BITS) - 1);
510 
511  if (ftype == FT_LONG) {
512  for (i = 0; i < tctx->avctx->channels; i++)
513  out[i] = (1./(1<<13)) *
514  mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
515  AMP_MAX, MULAW_MU);
516  } else {
517  for (i = 0; i < tctx->avctx->channels; i++) {
518  float val = (1./(1<<23)) *
519  mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
520  AMP_MAX, MULAW_MU);
521 
522  for (j = 0; j < sub; j++) {
523  out[i*sub + j] =
524  val*mulawinv(sub_step* 0.5 +
525  sub_step* get_bits(gb, SUB_GAIN_BITS),
527  }
528  }
529  }
530 }
531 
538 static void rearrange_lsp(int order, float *lsp, float min_dist)
539 {
540  int i;
541  float min_dist2 = min_dist * 0.5;
542  for (i = 1; i < order; i++)
543  if (lsp[i] - lsp[i-1] < min_dist) {
544  float avg = (lsp[i] + lsp[i-1]) * 0.5;
545 
546  lsp[i-1] = avg - min_dist2;
547  lsp[i ] = avg + min_dist2;
548  }
549 }
550 
551 static void decode_lsp(TwinContext *tctx, int lpc_idx1, uint8_t *lpc_idx2,
552  int lpc_hist_idx, float *lsp, float *hist)
553 {
554  const ModeTab *mtab = tctx->mtab;
555  int i, j;
556 
557  const float *cb = mtab->lspcodebook;
558  const float *cb2 = cb + (1 << mtab->lsp_bit1)*mtab->n_lsp;
559  const float *cb3 = cb2 + (1 << mtab->lsp_bit2)*mtab->n_lsp;
560 
561  const int8_t funny_rounding[4] = {
562  -2,
563  mtab->lsp_split == 4 ? -2 : 1,
564  mtab->lsp_split == 4 ? -2 : 1,
565  0
566  };
567 
568  j = 0;
569  for (i = 0; i < mtab->lsp_split; i++) {
570  int chunk_end = ((i + 1)*mtab->n_lsp + funny_rounding[i])/mtab->lsp_split;
571  for (; j < chunk_end; j++)
572  lsp[j] = cb [lpc_idx1 * mtab->n_lsp + j] +
573  cb2[lpc_idx2[i] * mtab->n_lsp + j];
574  }
575 
576  rearrange_lsp(mtab->n_lsp, lsp, 0.0001);
577 
578  for (i = 0; i < mtab->n_lsp; i++) {
579  float tmp1 = 1. - cb3[lpc_hist_idx*mtab->n_lsp + i];
580  float tmp2 = hist[i] * cb3[lpc_hist_idx*mtab->n_lsp + i];
581  hist[i] = lsp[i];
582  lsp[i] = lsp[i] * tmp1 + tmp2;
583  }
584 
585  rearrange_lsp(mtab->n_lsp, lsp, 0.0001);
586  rearrange_lsp(mtab->n_lsp, lsp, 0.000095);
588 }
589 
590 static void dec_lpc_spectrum_inv(TwinContext *tctx, float *lsp,
591  enum FrameType ftype, float *lpc)
592 {
593  int i;
594  int size = tctx->mtab->size / tctx->mtab->fmode[ftype].sub;
595 
596  for (i = 0; i < tctx->mtab->n_lsp; i++)
597  lsp[i] = 2*cos(lsp[i]);
598 
599  switch (ftype) {
600  case FT_LONG:
601  eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 8);
602  break;
603  case FT_MEDIUM:
604  eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 2);
605  break;
606  case FT_SHORT:
607  eval_lpcenv(tctx, lsp, lpc);
608  break;
609  }
610 }
611 
612 static void imdct_and_window(TwinContext *tctx, enum FrameType ftype, int wtype,
613  float *in, float *prev, int ch)
614 {
615  FFTContext *mdct = &tctx->mdct_ctx[ftype];
616  const ModeTab *mtab = tctx->mtab;
617  int bsize = mtab->size / mtab->fmode[ftype].sub;
618  int size = mtab->size;
619  float *buf1 = tctx->tmp_buf;
620  int j;
621  int wsize; // Window size
622  float *out = tctx->curr_frame + 2*ch*mtab->size;
623  float *out2 = out;
624  float *prev_buf;
625  int first_wsize;
626 
627  static const uint8_t wtype_to_wsize[] = {0, 0, 2, 2, 2, 1, 0, 1, 1};
628  int types_sizes[] = {
629  mtab->size / mtab->fmode[FT_LONG ].sub,
630  mtab->size / mtab->fmode[FT_MEDIUM].sub,
631  mtab->size / (2*mtab->fmode[FT_SHORT ].sub),
632  };
633 
634  wsize = types_sizes[wtype_to_wsize[wtype]];
635  first_wsize = wsize;
636  prev_buf = prev + (size - bsize)/2;
637 
638  for (j = 0; j < mtab->fmode[ftype].sub; j++) {
639  int sub_wtype = ftype == FT_MEDIUM ? 8 : wtype;
640 
641  if (!j && wtype == 4)
642  sub_wtype = 4;
643  else if (j == mtab->fmode[ftype].sub-1 && wtype == 7)
644  sub_wtype = 7;
645 
646  wsize = types_sizes[wtype_to_wsize[sub_wtype]];
647 
648  mdct->imdct_half(mdct, buf1 + bsize*j, in + bsize*j);
649 
650  tctx->dsp.vector_fmul_window(out2,
651  prev_buf + (bsize-wsize)/2,
652  buf1 + bsize*j,
653  ff_sine_windows[av_log2(wsize)],
654  wsize/2);
655  out2 += wsize;
656 
657  memcpy(out2, buf1 + bsize*j + wsize/2, (bsize - wsize/2)*sizeof(float));
658 
659  out2 += ftype == FT_MEDIUM ? (bsize-wsize)/2 : bsize - wsize;
660 
661  prev_buf = buf1 + bsize*j + bsize/2;
662  }
663 
664  tctx->last_block_pos[ch] = (size + first_wsize)/2;
665 }
666 
667 static void imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype,
668  float *out)
669 {
670  const ModeTab *mtab = tctx->mtab;
671  int size1, size2;
672  float *prev_buf = tctx->prev_frame + tctx->last_block_pos[0];
673  int i;
674 
675  for (i = 0; i < tctx->avctx->channels; i++) {
676  imdct_and_window(tctx, ftype, wtype,
677  tctx->spectrum + i*mtab->size,
678  prev_buf + 2*i*mtab->size,
679  i);
680  }
681 
682  if (!out)
683  return;
684 
685  size2 = tctx->last_block_pos[0];
686  size1 = mtab->size - size2;
687  if (tctx->avctx->channels == 2) {
688  tctx->dsp.butterflies_float_interleave(out, prev_buf,
689  &prev_buf[2*mtab->size],
690  size1);
691 
692  out += 2 * size1;
693 
695  &tctx->curr_frame[2*mtab->size],
696  size2);
697  } else {
698  memcpy(out, prev_buf, size1 * sizeof(*out));
699 
700  out += size1;
701 
702  memcpy(out, tctx->curr_frame, size2 * sizeof(*out));
703  }
704 
705 }
706 
707 static void dec_bark_env(TwinContext *tctx, const uint8_t *in, int use_hist,
708  int ch, float *out, float gain, enum FrameType ftype)
709 {
710  const ModeTab *mtab = tctx->mtab;
711  int i,j;
712  float *hist = tctx->bark_hist[ftype][ch];
713  float val = ((const float []) {0.4, 0.35, 0.28})[ftype];
714  int bark_n_coef = mtab->fmode[ftype].bark_n_coef;
715  int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef;
716  int idx = 0;
717 
718  for (i = 0; i < fw_cb_len; i++)
719  for (j = 0; j < bark_n_coef; j++, idx++) {
720  float tmp2 =
721  mtab->fmode[ftype].bark_cb[fw_cb_len*in[j] + i] * (1./4096);
722  float st = use_hist ?
723  (1. - val) * tmp2 + val*hist[idx] + 1. : tmp2 + 1.;
724 
725  hist[idx] = tmp2;
726  if (st < -1.) st = 1.;
727 
728  memset_float(out, st * gain, mtab->fmode[ftype].bark_tab[idx]);
729  out += mtab->fmode[ftype].bark_tab[idx];
730  }
731 
732 }
733 
735  float *out, enum FrameType ftype)
736 {
737  const ModeTab *mtab = tctx->mtab;
738  int channels = tctx->avctx->channels;
739  int sub = mtab->fmode[ftype].sub;
740  int block_size = mtab->size / sub;
741  float gain[CHANNELS_MAX*SUBBLOCKS_MAX];
742  float ppc_shape[PPC_SHAPE_LEN_MAX * CHANNELS_MAX * 4];
743  uint8_t bark1[CHANNELS_MAX][SUBBLOCKS_MAX][BARK_N_COEF_MAX];
744  uint8_t bark_use_hist[CHANNELS_MAX][SUBBLOCKS_MAX];
745 
746  uint8_t lpc_idx1[CHANNELS_MAX];
747  uint8_t lpc_idx2[CHANNELS_MAX][LSP_SPLIT_MAX];
748  uint8_t lpc_hist_idx[CHANNELS_MAX];
749 
750  int i, j, k;
751 
752  dequant(tctx, gb, out, ftype,
753  mtab->fmode[ftype].cb0, mtab->fmode[ftype].cb1,
754  mtab->fmode[ftype].cb_len_read);
755 
756  for (i = 0; i < channels; i++)
757  for (j = 0; j < sub; j++)
758  for (k = 0; k < mtab->fmode[ftype].bark_n_coef; k++)
759  bark1[i][j][k] =
760  get_bits(gb, mtab->fmode[ftype].bark_n_bit);
761 
762  for (i = 0; i < channels; i++)
763  for (j = 0; j < sub; j++)
764  bark_use_hist[i][j] = get_bits1(gb);
765 
766  dec_gain(tctx, gb, ftype, gain);
767 
768  for (i = 0; i < channels; i++) {
769  lpc_hist_idx[i] = get_bits(gb, tctx->mtab->lsp_bit0);
770  lpc_idx1 [i] = get_bits(gb, tctx->mtab->lsp_bit1);
771 
772  for (j = 0; j < tctx->mtab->lsp_split; j++)
773  lpc_idx2[i][j] = get_bits(gb, tctx->mtab->lsp_bit2);
774  }
775 
776  if (ftype == FT_LONG) {
777  int cb_len_p = (tctx->n_div[3] + mtab->ppc_shape_len*channels - 1)/
778  tctx->n_div[3];
779  dequant(tctx, gb, ppc_shape, FT_PPC, mtab->ppc_shape_cb,
780  mtab->ppc_shape_cb + cb_len_p*PPC_SHAPE_CB_SIZE, cb_len_p);
781  }
782 
783  for (i = 0; i < channels; i++) {
784  float *chunk = out + mtab->size * i;
785  float lsp[LSP_COEFS_MAX];
786 
787  for (j = 0; j < sub; j++) {
788  dec_bark_env(tctx, bark1[i][j], bark_use_hist[i][j], i,
789  tctx->tmp_buf, gain[sub*i+j], ftype);
790 
791  tctx->dsp.vector_fmul(chunk + block_size*j, chunk + block_size*j, tctx->tmp_buf,
792  block_size);
793 
794  }
795 
796  if (ftype == FT_LONG) {
797  float pgain_step = 25000. / ((1 << mtab->pgain_bit) - 1);
798  int p_coef = get_bits(gb, tctx->mtab->ppc_period_bit);
799  int g_coef = get_bits(gb, tctx->mtab->pgain_bit);
800  float v = 1./8192*
801  mulawinv(pgain_step*g_coef+ pgain_step/2, 25000., PGAIN_MU);
802 
803  decode_ppc(tctx, p_coef, ppc_shape + i*mtab->ppc_shape_len, v,
804  chunk);
805  }
806 
807  decode_lsp(tctx, lpc_idx1[i], lpc_idx2[i], lpc_hist_idx[i], lsp,
808  tctx->lsp_hist[i]);
809 
810  dec_lpc_spectrum_inv(tctx, lsp, ftype, tctx->tmp_buf);
811 
812  for (j = 0; j < mtab->fmode[ftype].sub; j++) {
813  tctx->dsp.vector_fmul(chunk, chunk, tctx->tmp_buf, block_size);
814  chunk += block_size;
815  }
816  }
817 }
818 
819 static int twin_decode_frame(AVCodecContext * avctx, void *data,
820  int *got_frame_ptr, AVPacket *avpkt)
821 {
822  const uint8_t *buf = avpkt->data;
823  int buf_size = avpkt->size;
824  TwinContext *tctx = avctx->priv_data;
825  GetBitContext gb;
826  const ModeTab *mtab = tctx->mtab;
827  float *out = NULL;
828  enum FrameType ftype;
829  int window_type, ret;
830  static const enum FrameType wtype_to_ftype_table[] = {
832  FT_MEDIUM, FT_LONG, FT_LONG, FT_MEDIUM, FT_MEDIUM
833  };
834 
835  if (buf_size*8 < avctx->bit_rate*mtab->size/avctx->sample_rate + 8) {
836  av_log(avctx, AV_LOG_ERROR,
837  "Frame too small (%d bytes). Truncated file?\n", buf_size);
838  return AVERROR(EINVAL);
839  }
840 
841  /* get output buffer */
842  if (tctx->discarded_packets >= 2) {
843  tctx->frame.nb_samples = mtab->size;
844  if ((ret = ff_get_buffer(avctx, &tctx->frame)) < 0) {
845  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
846  return ret;
847  }
848  out = (float *)tctx->frame.data[0];
849  }
850 
851  init_get_bits(&gb, buf, buf_size * 8);
852  skip_bits(&gb, get_bits(&gb, 8));
853  window_type = get_bits(&gb, WINDOW_TYPE_BITS);
854 
855  if (window_type > 8) {
856  av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
857  return -1;
858  }
859 
860  ftype = wtype_to_ftype_table[window_type];
861 
862  read_and_decode_spectrum(tctx, &gb, tctx->spectrum, ftype);
863 
864  imdct_output(tctx, ftype, window_type, out);
865 
866  FFSWAP(float*, tctx->curr_frame, tctx->prev_frame);
867 
868  if (tctx->discarded_packets < 2) {
869  tctx->discarded_packets++;
870  *got_frame_ptr = 0;
871  return buf_size;
872  }
873 
874  *got_frame_ptr = 1;
875  *(AVFrame *)data = tctx->frame;;
876 
877  return buf_size;
878 }
879 
884 {
885  int i, j, ret;
886  const ModeTab *mtab = tctx->mtab;
887  int size_s = mtab->size / mtab->fmode[FT_SHORT].sub;
888  int size_m = mtab->size / mtab->fmode[FT_MEDIUM].sub;
889  int channels = tctx->avctx->channels;
890  float norm = channels == 1 ? 2. : 1.;
891 
892  for (i = 0; i < 3; i++) {
893  int bsize = tctx->mtab->size/tctx->mtab->fmode[i].sub;
894  if ((ret = ff_mdct_init(&tctx->mdct_ctx[i], av_log2(bsize) + 1, 1,
895  -sqrt(norm/bsize) / (1<<15))))
896  return ret;
897  }
898 
899  FF_ALLOC_OR_GOTO(tctx->avctx, tctx->tmp_buf,
900  mtab->size * sizeof(*tctx->tmp_buf), alloc_fail);
901 
902  FF_ALLOC_OR_GOTO(tctx->avctx, tctx->spectrum,
903  2 * mtab->size * channels * sizeof(*tctx->spectrum),
904  alloc_fail);
905  FF_ALLOC_OR_GOTO(tctx->avctx, tctx->curr_frame,
906  2 * mtab->size * channels * sizeof(*tctx->curr_frame),
907  alloc_fail);
908  FF_ALLOC_OR_GOTO(tctx->avctx, tctx->prev_frame,
909  2 * mtab->size * channels * sizeof(*tctx->prev_frame),
910  alloc_fail);
911 
912  for (i = 0; i < 3; i++) {
913  int m = 4*mtab->size/mtab->fmode[i].sub;
914  double freq = 2*M_PI/m;
915  FF_ALLOC_OR_GOTO(tctx->avctx, tctx->cos_tabs[i],
916  (m / 4) * sizeof(*tctx->cos_tabs[i]), alloc_fail);
917 
918  for (j = 0; j <= m/8; j++)
919  tctx->cos_tabs[i][j] = cos((2*j + 1)*freq);
920  for (j = 1; j < m/8; j++)
921  tctx->cos_tabs[i][m/4-j] = tctx->cos_tabs[i][j];
922  }
923 
924 
925  ff_init_ff_sine_windows(av_log2(size_m));
926  ff_init_ff_sine_windows(av_log2(size_s/2));
927  ff_init_ff_sine_windows(av_log2(mtab->size));
928 
929  return 0;
930 alloc_fail:
931  return AVERROR(ENOMEM);
932 }
933 
940 static void permutate_in_line(int16_t *tab, int num_vect, int num_blocks,
941  int block_size,
942  const uint8_t line_len[2], int length_div,
943  enum FrameType ftype)
944 
945 {
946  int i,j;
947 
948  for (i = 0; i < line_len[0]; i++) {
949  int shift;
950 
951  if (num_blocks == 1 ||
952  (ftype == FT_LONG && num_vect % num_blocks) ||
953  (ftype != FT_LONG && num_vect & 1 ) ||
954  i == line_len[1]) {
955  shift = 0;
956  } else if (ftype == FT_LONG) {
957  shift = i;
958  } else
959  shift = i*i;
960 
961  for (j = 0; j < num_vect && (j+num_vect*i < block_size*num_blocks); j++)
962  tab[i*num_vect+j] = i*num_vect + (j + shift) % num_vect;
963  }
964 }
965 
981 static void transpose_perm(int16_t *out, int16_t *in, int num_vect,
982  const uint8_t line_len[2], int length_div)
983 {
984  int i,j;
985  int cont= 0;
986  for (i = 0; i < num_vect; i++)
987  for (j = 0; j < line_len[i >= length_div]; j++)
988  out[cont++] = in[j*num_vect + i];
989 }
990 
991 static void linear_perm(int16_t *out, int16_t *in, int n_blocks, int size)
992 {
993  int block_size = size/n_blocks;
994  int i;
995 
996  for (i = 0; i < size; i++)
997  out[i] = block_size * (in[i] % n_blocks) + in[i] / n_blocks;
998 }
999 
1000 static av_cold void construct_perm_table(TwinContext *tctx, int ftype)
1001 {
1002  int block_size;
1003  const ModeTab *mtab = tctx->mtab;
1004  int size;
1005  int16_t *tmp_perm = (int16_t *) tctx->tmp_buf;
1006 
1007  if (ftype == FT_PPC) {
1008  size = tctx->avctx->channels;
1009  block_size = mtab->ppc_shape_len;
1010  } else {
1011  size = tctx->avctx->channels * mtab->fmode[ftype].sub;
1012  block_size = mtab->size / mtab->fmode[ftype].sub;
1013  }
1014 
1015  permutate_in_line(tmp_perm, tctx->n_div[ftype], size,
1016  block_size, tctx->length[ftype],
1017  tctx->length_change[ftype], ftype);
1018 
1019  transpose_perm(tctx->permut[ftype], tmp_perm, tctx->n_div[ftype],
1020  tctx->length[ftype], tctx->length_change[ftype]);
1021 
1022  linear_perm(tctx->permut[ftype], tctx->permut[ftype], size,
1023  size*block_size);
1024 }
1025 
1027 {
1028  const ModeTab *mtab = tctx->mtab;
1029  int n_ch = tctx->avctx->channels;
1030  int total_fr_bits = tctx->avctx->bit_rate*mtab->size/
1031  tctx->avctx->sample_rate;
1032 
1033  int lsp_bits_per_block = n_ch*(mtab->lsp_bit0 + mtab->lsp_bit1 +
1034  mtab->lsp_split*mtab->lsp_bit2);
1035 
1036  int ppc_bits = n_ch*(mtab->pgain_bit + mtab->ppc_shape_bit +
1037  mtab->ppc_period_bit);
1038 
1039  int bsize_no_main_cb[3];
1040  int bse_bits[3];
1041  int i;
1042  enum FrameType frametype;
1043 
1044  for (i = 0; i < 3; i++)
1045  // +1 for history usage switch
1046  bse_bits[i] = n_ch *
1047  (mtab->fmode[i].bark_n_coef * mtab->fmode[i].bark_n_bit + 1);
1048 
1049  bsize_no_main_cb[2] = bse_bits[2] + lsp_bits_per_block + ppc_bits +
1050  WINDOW_TYPE_BITS + n_ch*GAIN_BITS;
1051 
1052  for (i = 0; i < 2; i++)
1053  bsize_no_main_cb[i] =
1054  lsp_bits_per_block + n_ch*GAIN_BITS + WINDOW_TYPE_BITS +
1055  mtab->fmode[i].sub*(bse_bits[i] + n_ch*SUB_GAIN_BITS);
1056 
1057  // The remaining bits are all used for the main spectrum coefficients
1058  for (i = 0; i < 4; i++) {
1059  int bit_size;
1060  int vect_size;
1061  int rounded_up, rounded_down, num_rounded_down, num_rounded_up;
1062  if (i == 3) {
1063  bit_size = n_ch * mtab->ppc_shape_bit;
1064  vect_size = n_ch * mtab->ppc_shape_len;
1065  } else {
1066  bit_size = total_fr_bits - bsize_no_main_cb[i];
1067  vect_size = n_ch * mtab->size;
1068  }
1069 
1070  tctx->n_div[i] = (bit_size + 13) / 14;
1071 
1072  rounded_up = (bit_size + tctx->n_div[i] - 1)/tctx->n_div[i];
1073  rounded_down = (bit_size )/tctx->n_div[i];
1074  num_rounded_down = rounded_up * tctx->n_div[i] - bit_size;
1075  num_rounded_up = tctx->n_div[i] - num_rounded_down;
1076  tctx->bits_main_spec[0][i][0] = (rounded_up + 1)/2;
1077  tctx->bits_main_spec[1][i][0] = (rounded_up )/2;
1078  tctx->bits_main_spec[0][i][1] = (rounded_down + 1)/2;
1079  tctx->bits_main_spec[1][i][1] = (rounded_down )/2;
1080  tctx->bits_main_spec_change[i] = num_rounded_up;
1081 
1082  rounded_up = (vect_size + tctx->n_div[i] - 1)/tctx->n_div[i];
1083  rounded_down = (vect_size )/tctx->n_div[i];
1084  num_rounded_down = rounded_up * tctx->n_div[i] - vect_size;
1085  num_rounded_up = tctx->n_div[i] - num_rounded_down;
1086  tctx->length[i][0] = rounded_up;
1087  tctx->length[i][1] = rounded_down;
1088  tctx->length_change[i] = num_rounded_up;
1089  }
1090 
1091  for (frametype = FT_SHORT; frametype <= FT_PPC; frametype++)
1092  construct_perm_table(tctx, frametype);
1093 }
1094 
1096 {
1097  TwinContext *tctx = avctx->priv_data;
1098  int i;
1099 
1100  for (i = 0; i < 3; i++) {
1101  ff_mdct_end(&tctx->mdct_ctx[i]);
1102  av_free(tctx->cos_tabs[i]);
1103  }
1104 
1105 
1106  av_free(tctx->curr_frame);
1107  av_free(tctx->spectrum);
1108  av_free(tctx->prev_frame);
1109  av_free(tctx->tmp_buf);
1110 
1111  return 0;
1112 }
1113 
1115 {
1116  int ret;
1117  TwinContext *tctx = avctx->priv_data;
1118  int isampf, ibps;
1119 
1120  tctx->avctx = avctx;
1121  avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
1122 
1123  if (!avctx->extradata || avctx->extradata_size < 12) {
1124  av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n");
1125  return AVERROR_INVALIDDATA;
1126  }
1127  avctx->channels = AV_RB32(avctx->extradata ) + 1;
1128  avctx->bit_rate = AV_RB32(avctx->extradata + 4) * 1000;
1129  isampf = AV_RB32(avctx->extradata + 8);
1130  switch (isampf) {
1131  case 44: avctx->sample_rate = 44100; break;
1132  case 22: avctx->sample_rate = 22050; break;
1133  case 11: avctx->sample_rate = 11025; break;
1134  default: avctx->sample_rate = isampf * 1000; break;
1135  }
1136 
1137  if (avctx->channels > CHANNELS_MAX) {
1138  av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n",
1139  avctx->channels);
1140  return -1;
1141  }
1142  ibps = avctx->bit_rate / (1000 * avctx->channels);
1143  if (ibps < 8 || ibps > 48) {
1144  av_log(avctx, AV_LOG_ERROR, "Bad bitrate per channel value %d\n", ibps);
1145  return AVERROR_INVALIDDATA;
1146  }
1147 
1148  switch ((isampf << 8) + ibps) {
1149  case (8 <<8) + 8: tctx->mtab = &mode_08_08; break;
1150  case (11<<8) + 8: tctx->mtab = &mode_11_08; break;
1151  case (11<<8) + 10: tctx->mtab = &mode_11_10; break;
1152  case (16<<8) + 16: tctx->mtab = &mode_16_16; break;
1153  case (22<<8) + 20: tctx->mtab = &mode_22_20; break;
1154  case (22<<8) + 24: tctx->mtab = &mode_22_24; break;
1155  case (22<<8) + 32: tctx->mtab = &mode_22_32; break;
1156  case (44<<8) + 40: tctx->mtab = &mode_44_40; break;
1157  case (44<<8) + 48: tctx->mtab = &mode_44_48; break;
1158  default:
1159  av_log(avctx, AV_LOG_ERROR, "This version does not support %d kHz - %d kbit/s/ch mode.\n", isampf, isampf);
1160  return -1;
1161  }
1162 
1163  dsputil_init(&tctx->dsp, avctx);
1164  if ((ret = init_mdct_win(tctx))) {
1165  av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
1166  twin_decode_close(avctx);
1167  return ret;
1168  }
1169  init_bitstream_params(tctx);
1170 
1171  memset_float(tctx->bark_hist[0][0], 0.1, FF_ARRAY_ELEMS(tctx->bark_hist));
1172 
1174  avctx->coded_frame = &tctx->frame;
1175 
1176  return 0;
1177 }
1178 
1180  .name = "twinvq",
1181  .type = AVMEDIA_TYPE_AUDIO,
1182  .id = CODEC_ID_TWINVQ,
1183  .priv_data_size = sizeof(TwinContext),
1187  .capabilities = CODEC_CAP_DR1,
1188  .long_name = NULL_IF_CONFIG_SMALL("VQF TwinVQ"),
1189 };