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 
23 #include "libavutil/float_dsp.h"
24 #include "avcodec.h"
25 #include "get_bits.h"
26 #include "dsputil.h"
27 #include "fft.h"
28 #include "internal.h"
29 #include "lsp.h"
30 #include "sinewin.h"
31 
32 #include <math.h>
33 #include <stdint.h>
34 
35 #include "twinvq_data.h"
36 
37 enum FrameType {
38  FT_SHORT = 0,
42 };
43 
47 struct FrameMode {
49  const uint16_t *bark_tab;
50 
53 
54  const int16_t *bark_cb;
57 
59 
60  const int16_t *cb0;
61  const int16_t *cb1;
63 
65 };
66 
71 typedef struct {
72  struct FrameMode fmode[3];
73 
74  uint16_t size;
76  const float *lspcodebook;
77 
78  /* number of bits of the different LSP CB coefficients */
82 
84  const int16_t *ppc_shape_cb;
85 
88 
92 
94  uint16_t peak_per2wid;
95 } ModeTab;
96 
97 static const ModeTab mode_08_08 = {
98  {
99  { 8, bark_tab_s08_64, 10, tab.fcb08s , 1, 5, tab.cb0808s0, tab.cb0808s1, 18},
100  { 2, bark_tab_m08_256, 20, tab.fcb08m , 2, 5, tab.cb0808m0, tab.cb0808m1, 16},
101  { 1, bark_tab_l08_512, 30, tab.fcb08l , 3, 6, tab.cb0808l0, tab.cb0808l1, 17}
102  },
103  512 , 12, tab.lsp08, 1, 5, 3, 3, tab.shape08 , 8, 28, 20, 6, 40
104 };
105 
106 static const ModeTab mode_11_08 = {
107  {
108  { 8, bark_tab_s11_64, 10, tab.fcb11s , 1, 5, tab.cb1108s0, tab.cb1108s1, 29},
109  { 2, bark_tab_m11_256, 20, tab.fcb11m , 2, 5, tab.cb1108m0, tab.cb1108m1, 24},
110  { 1, bark_tab_l11_512, 30, tab.fcb11l , 3, 6, tab.cb1108l0, tab.cb1108l1, 27}
111  },
112  512 , 16, tab.lsp11, 1, 6, 4, 3, tab.shape11 , 9, 36, 30, 7, 90
113 };
114 
115 static const ModeTab mode_11_10 = {
116  {
117  { 8, bark_tab_s11_64, 10, tab.fcb11s , 1, 5, tab.cb1110s0, tab.cb1110s1, 21},
118  { 2, bark_tab_m11_256, 20, tab.fcb11m , 2, 5, tab.cb1110m0, tab.cb1110m1, 18},
119  { 1, bark_tab_l11_512, 30, tab.fcb11l , 3, 6, tab.cb1110l0, tab.cb1110l1, 20}
120  },
121  512 , 16, tab.lsp11, 1, 6, 4, 3, tab.shape11 , 9, 36, 30, 7, 90
122 };
123 
124 static const ModeTab mode_16_16 = {
125  {
126  { 8, bark_tab_s16_128, 10, tab.fcb16s , 1, 5, tab.cb1616s0, tab.cb1616s1, 16},
127  { 2, bark_tab_m16_512, 20, tab.fcb16m , 2, 5, tab.cb1616m0, tab.cb1616m1, 15},
128  { 1, bark_tab_l16_1024,30, tab.fcb16l , 3, 6, tab.cb1616l0, tab.cb1616l1, 16}
129  },
130  1024, 16, tab.lsp16, 1, 6, 4, 3, tab.shape16 , 9, 56, 60, 7, 180
131 };
132 
133 static const ModeTab mode_22_20 = {
134  {
135  { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2220s0, tab.cb2220s1, 18},
136  { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2220m0, tab.cb2220m1, 17},
137  { 1, bark_tab_l22_1024,32, tab.fcb22l_1, 4, 6, tab.cb2220l0, tab.cb2220l1, 18}
138  },
139  1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
140 };
141 
142 static const ModeTab mode_22_24 = {
143  {
144  { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2224s0, tab.cb2224s1, 15},
145  { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2224m0, tab.cb2224m1, 14},
146  { 1, bark_tab_l22_1024,32, tab.fcb22l_1, 4, 6, tab.cb2224l0, tab.cb2224l1, 15}
147  },
148  1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
149 };
150 
151 static const ModeTab mode_22_32 = {
152  {
153  { 4, bark_tab_s22_128, 10, tab.fcb22s_2, 1, 6, tab.cb2232s0, tab.cb2232s1, 11},
154  { 2, bark_tab_m22_256, 20, tab.fcb22m_2, 2, 6, tab.cb2232m0, tab.cb2232m1, 11},
155  { 1, bark_tab_l22_512, 32, tab.fcb22l_2, 4, 6, tab.cb2232l0, tab.cb2232l1, 12}
156  },
157  512 , 16, tab.lsp22_2, 1, 6, 4, 4, tab.shape22_2, 9, 56, 36, 7, 72
158 };
159 
160 static const ModeTab mode_44_40 = {
161  {
162  {16, bark_tab_s44_128, 10, tab.fcb44s , 1, 6, tab.cb4440s0, tab.cb4440s1, 18},
163  { 4, bark_tab_m44_512, 20, tab.fcb44m , 2, 6, tab.cb4440m0, tab.cb4440m1, 17},
164  { 1, bark_tab_l44_2048,40, tab.fcb44l , 4, 6, tab.cb4440l0, tab.cb4440l1, 17}
165  },
166  2048, 20, tab.lsp44, 1, 6, 4, 4, tab.shape44 , 9, 84, 54, 7, 432
167 };
168 
169 static const ModeTab mode_44_48 = {
170  {
171  {16, bark_tab_s44_128, 10, tab.fcb44s , 1, 6, tab.cb4448s0, tab.cb4448s1, 15},
172  { 4, bark_tab_m44_512, 20, tab.fcb44m , 2, 6, tab.cb4448m0, tab.cb4448m1, 14},
173  { 1, bark_tab_l44_2048,40, tab.fcb44l , 4, 6, tab.cb4448l0, tab.cb4448l1, 14}
174  },
175  2048, 20, tab.lsp44, 1, 6, 4, 4, tab.shape44 , 9, 84, 54, 7, 432
176 };
177 
178 typedef struct TwinContext {
183 
184  const ModeTab *mtab;
185 
186  // history
187  float lsp_hist[2][20];
188  float bark_hist[3][2][40];
189 
190  // bitstream parameters
191  int16_t permut[4][4096];
192  uint8_t length[4][2];
196  int n_div[4];
197 
198  float *spectrum;
199  float *curr_frame;
200  float *prev_frame;
203 
204  float *cos_tabs[3];
205 
206  // scratch buffers
207  float *tmp_buf;
208 } TwinContext;
209 
210 #define PPC_SHAPE_CB_SIZE 64
211 #define PPC_SHAPE_LEN_MAX 60
212 #define SUB_AMP_MAX 4500.0
213 #define MULAW_MU 100.0
214 #define GAIN_BITS 8
215 #define AMP_MAX 13000.0
216 #define SUB_GAIN_BITS 5
217 #define WINDOW_TYPE_BITS 4
218 #define PGAIN_MU 200
219 #define LSP_COEFS_MAX 20
220 #define LSP_SPLIT_MAX 4
221 #define CHANNELS_MAX 2
222 #define SUBBLOCKS_MAX 16
223 #define BARK_N_COEF_MAX 4
224 
226 static void memset_float(float *buf, float val, int size)
227 {
228  while (size--)
229  *buf++ = val;
230 }
231 
244 static float eval_lpc_spectrum(const float *lsp, float cos_val, int order)
245 {
246  int j;
247  float p = 0.5f;
248  float q = 0.5f;
249  float two_cos_w = 2.0f*cos_val;
250 
251  for (j = 0; j + 1 < order; j += 2*2) {
252  // Unroll the loop once since order is a multiple of four
253  q *= lsp[j ] - two_cos_w;
254  p *= lsp[j+1] - two_cos_w;
255 
256  q *= lsp[j+2] - two_cos_w;
257  p *= lsp[j+3] - two_cos_w;
258  }
259 
260  p *= p * (2.0f - two_cos_w);
261  q *= q * (2.0f + two_cos_w);
262 
263  return 0.5 / (p + q);
264 }
265 
269 static void eval_lpcenv(TwinContext *tctx, const float *cos_vals, float *lpc)
270 {
271  int i;
272  const ModeTab *mtab = tctx->mtab;
273  int size_s = mtab->size / mtab->fmode[FT_SHORT].sub;
274 
275  for (i = 0; i < size_s/2; i++) {
276  float cos_i = tctx->cos_tabs[0][i];
277  lpc[i] = eval_lpc_spectrum(cos_vals, cos_i, mtab->n_lsp);
278  lpc[size_s-i-1] = eval_lpc_spectrum(cos_vals, -cos_i, mtab->n_lsp);
279  }
280 }
281 
282 static void interpolate(float *out, float v1, float v2, int size)
283 {
284  int i;
285  float step = (v1 - v2)/(size + 1);
286 
287  for (i = 0; i < size; i++) {
288  v2 += step;
289  out[i] = v2;
290  }
291 }
292 
293 static inline float get_cos(int idx, int part, const float *cos_tab, int size)
294 {
295  return part ? -cos_tab[size - idx - 1] :
296  cos_tab[ idx ];
297 }
298 
313 static inline void eval_lpcenv_or_interp(TwinContext *tctx,
314  enum FrameType ftype,
315  float *out, const float *in,
316  int size, int step, int part)
317 {
318  int i;
319  const ModeTab *mtab = tctx->mtab;
320  const float *cos_tab = tctx->cos_tabs[ftype];
321 
322  // Fill the 's'
323  for (i = 0; i < size; i += step)
324  out[i] =
326  get_cos(i, part, cos_tab, size),
327  mtab->n_lsp);
328 
329  // Fill the 'iiiibiiii'
330  for (i = step; i <= size - 2*step; i += step) {
331  if (out[i + step] + out[i - step] > 1.95*out[i] ||
332  out[i + step] >= out[i - step]) {
333  interpolate(out + i - step + 1, out[i], out[i-step], step - 1);
334  } else {
335  out[i - step/2] =
337  get_cos(i-step/2, part, cos_tab, size),
338  mtab->n_lsp);
339  interpolate(out + i - step + 1, out[i-step/2], out[i-step ], step/2 - 1);
340  interpolate(out + i - step/2 + 1, out[i ], out[i-step/2], step/2 - 1);
341  }
342  }
343 
344  interpolate(out + size - 2*step + 1, out[size-step], out[size - 2*step], step - 1);
345 }
346 
347 static void eval_lpcenv_2parts(TwinContext *tctx, enum FrameType ftype,
348  const float *buf, float *lpc,
349  int size, int step)
350 {
351  eval_lpcenv_or_interp(tctx, ftype, lpc , buf, size/2, step, 0);
352  eval_lpcenv_or_interp(tctx, ftype, lpc + size/2, buf, size/2, 2*step, 1);
353 
354  interpolate(lpc+size/2-step+1, lpc[size/2], lpc[size/2-step], step);
355 
356  memset_float(lpc + size - 2*step + 1, lpc[size - 2*step], 2*step - 1);
357 }
358 
364 static void dequant(TwinContext *tctx, GetBitContext *gb, float *out,
365  enum FrameType ftype,
366  const int16_t *cb0, const int16_t *cb1, int cb_len)
367 {
368  int pos = 0;
369  int i, j;
370 
371  for (i = 0; i < tctx->n_div[ftype]; i++) {
372  int tmp0, tmp1;
373  int sign0 = 1;
374  int sign1 = 1;
375  const int16_t *tab0, *tab1;
376  int length = tctx->length[ftype][i >= tctx->length_change[ftype]];
377  int bitstream_second_part = (i >= tctx->bits_main_spec_change[ftype]);
378 
379  int bits = tctx->bits_main_spec[0][ftype][bitstream_second_part];
380  if (bits == 7) {
381  if (get_bits1(gb))
382  sign0 = -1;
383  bits = 6;
384  }
385  tmp0 = get_bits(gb, bits);
386 
387  bits = tctx->bits_main_spec[1][ftype][bitstream_second_part];
388 
389  if (bits == 7) {
390  if (get_bits1(gb))
391  sign1 = -1;
392 
393  bits = 6;
394  }
395  tmp1 = get_bits(gb, bits);
396 
397  tab0 = cb0 + tmp0*cb_len;
398  tab1 = cb1 + tmp1*cb_len;
399 
400  for (j = 0; j < length; j++)
401  out[tctx->permut[ftype][pos+j]] = sign0*tab0[j] + sign1*tab1[j];
402 
403  pos += length;
404  }
405 
406 }
407 
408 static inline float mulawinv(float y, float clip, float mu)
409 {
410  y = av_clipf(y/clip, -1, 1);
411  return clip * FFSIGN(y) * (exp(log(1+mu) * fabs(y)) - 1) / mu;
412 }
413 
434 static int very_broken_op(int a, int b)
435 {
436  int x = a*b + 200;
437  int size;
438  const uint8_t *rtab;
439 
440  if (x%400 || b%5)
441  return x/400;
442 
443  x /= 400;
444 
445  size = tabs[b/5].size;
446  rtab = tabs[b/5].tab;
447  return x - rtab[size*av_log2(2*(x - 1)/size)+(x - 1)%size];
448 }
449 
455 static void add_peak(int period, int width, const float *shape,
456  float ppc_gain, float *speech, int len)
457 {
458  int i, j;
459 
460  const float *shape_end = shape + len;
461  int center;
462 
463  // First peak centered around zero
464  for (i = 0; i < width/2; i++)
465  speech[i] += ppc_gain * *shape++;
466 
467  for (i = 1; i < ROUNDED_DIV(len,width) ; i++) {
468  center = very_broken_op(period, i);
469  for (j = -width/2; j < (width+1)/2; j++)
470  speech[j+center] += ppc_gain * *shape++;
471  }
472 
473  // For the last block, be careful not to go beyond the end of the buffer
474  center = very_broken_op(period, i);
475  for (j = -width/2; j < (width + 1)/2 && shape < shape_end; j++)
476  speech[j+center] += ppc_gain * *shape++;
477 }
478 
479 static void decode_ppc(TwinContext *tctx, int period_coef, const float *shape,
480  float ppc_gain, float *speech)
481 {
482  const ModeTab *mtab = tctx->mtab;
483  int isampf = tctx->avctx->sample_rate/1000;
484  int ibps = tctx->avctx->bit_rate/(1000 * tctx->avctx->channels);
485  int min_period = ROUNDED_DIV( 40*2*mtab->size, isampf);
486  int max_period = ROUNDED_DIV(6*40*2*mtab->size, isampf);
487  int period_range = max_period - min_period;
488 
489  // This is actually the period multiplied by 400. It is just linearly coded
490  // between its maximum and minimum value.
491  int period = min_period +
492  ROUNDED_DIV(period_coef*period_range, (1 << mtab->ppc_period_bit) - 1);
493  int width;
494 
495  if (isampf == 22 && ibps == 32) {
496  // For some unknown reason, NTT decided to code this case differently...
497  width = ROUNDED_DIV((period + 800)* mtab->peak_per2wid, 400*mtab->size);
498  } else
499  width = (period )* mtab->peak_per2wid/(400*mtab->size);
500 
501  add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len);
502 }
503 
504 static void dec_gain(TwinContext *tctx, GetBitContext *gb, enum FrameType ftype,
505  float *out)
506 {
507  const ModeTab *mtab = tctx->mtab;
508  int i, j;
509  int sub = mtab->fmode[ftype].sub;
510  float step = AMP_MAX / ((1 << GAIN_BITS) - 1);
511  float sub_step = SUB_AMP_MAX / ((1 << SUB_GAIN_BITS) - 1);
512 
513  if (ftype == FT_LONG) {
514  for (i = 0; i < tctx->avctx->channels; i++)
515  out[i] = (1./(1<<13)) *
516  mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
517  AMP_MAX, MULAW_MU);
518  } else {
519  for (i = 0; i < tctx->avctx->channels; i++) {
520  float val = (1./(1<<23)) *
521  mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
522  AMP_MAX, MULAW_MU);
523 
524  for (j = 0; j < sub; j++) {
525  out[i*sub + j] =
526  val*mulawinv(sub_step* 0.5 +
527  sub_step* get_bits(gb, SUB_GAIN_BITS),
529  }
530  }
531  }
532 }
533 
540 static void rearrange_lsp(int order, float *lsp, float min_dist)
541 {
542  int i;
543  float min_dist2 = min_dist * 0.5;
544  for (i = 1; i < order; i++)
545  if (lsp[i] - lsp[i-1] < min_dist) {
546  float avg = (lsp[i] + lsp[i-1]) * 0.5;
547 
548  lsp[i-1] = avg - min_dist2;
549  lsp[i ] = avg + min_dist2;
550  }
551 }
552 
553 static void decode_lsp(TwinContext *tctx, int lpc_idx1, uint8_t *lpc_idx2,
554  int lpc_hist_idx, float *lsp, float *hist)
555 {
556  const ModeTab *mtab = tctx->mtab;
557  int i, j;
558 
559  const float *cb = mtab->lspcodebook;
560  const float *cb2 = cb + (1 << mtab->lsp_bit1)*mtab->n_lsp;
561  const float *cb3 = cb2 + (1 << mtab->lsp_bit2)*mtab->n_lsp;
562 
563  const int8_t funny_rounding[4] = {
564  -2,
565  mtab->lsp_split == 4 ? -2 : 1,
566  mtab->lsp_split == 4 ? -2 : 1,
567  0
568  };
569 
570  j = 0;
571  for (i = 0; i < mtab->lsp_split; i++) {
572  int chunk_end = ((i + 1)*mtab->n_lsp + funny_rounding[i])/mtab->lsp_split;
573  for (; j < chunk_end; j++)
574  lsp[j] = cb [lpc_idx1 * mtab->n_lsp + j] +
575  cb2[lpc_idx2[i] * mtab->n_lsp + j];
576  }
577 
578  rearrange_lsp(mtab->n_lsp, lsp, 0.0001);
579 
580  for (i = 0; i < mtab->n_lsp; i++) {
581  float tmp1 = 1. - cb3[lpc_hist_idx*mtab->n_lsp + i];
582  float tmp2 = hist[i] * cb3[lpc_hist_idx*mtab->n_lsp + i];
583  hist[i] = lsp[i];
584  lsp[i] = lsp[i] * tmp1 + tmp2;
585  }
586 
587  rearrange_lsp(mtab->n_lsp, lsp, 0.0001);
588  rearrange_lsp(mtab->n_lsp, lsp, 0.000095);
590 }
591 
592 static void dec_lpc_spectrum_inv(TwinContext *tctx, float *lsp,
593  enum FrameType ftype, float *lpc)
594 {
595  int i;
596  int size = tctx->mtab->size / tctx->mtab->fmode[ftype].sub;
597 
598  for (i = 0; i < tctx->mtab->n_lsp; i++)
599  lsp[i] = 2*cos(lsp[i]);
600 
601  switch (ftype) {
602  case FT_LONG:
603  eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 8);
604  break;
605  case FT_MEDIUM:
606  eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 2);
607  break;
608  case FT_SHORT:
609  eval_lpcenv(tctx, lsp, lpc);
610  break;
611  }
612 }
613 
614 static void imdct_and_window(TwinContext *tctx, enum FrameType ftype, int wtype,
615  float *in, float *prev, int ch)
616 {
617  FFTContext *mdct = &tctx->mdct_ctx[ftype];
618  const ModeTab *mtab = tctx->mtab;
619  int bsize = mtab->size / mtab->fmode[ftype].sub;
620  int size = mtab->size;
621  float *buf1 = tctx->tmp_buf;
622  int j;
623  int wsize; // Window size
624  float *out = tctx->curr_frame + 2*ch*mtab->size;
625  float *out2 = out;
626  float *prev_buf;
627  int first_wsize;
628 
629  static const uint8_t wtype_to_wsize[] = {0, 0, 2, 2, 2, 1, 0, 1, 1};
630  int types_sizes[] = {
631  mtab->size / mtab->fmode[FT_LONG ].sub,
632  mtab->size / mtab->fmode[FT_MEDIUM].sub,
633  mtab->size / (2*mtab->fmode[FT_SHORT ].sub),
634  };
635 
636  wsize = types_sizes[wtype_to_wsize[wtype]];
637  first_wsize = wsize;
638  prev_buf = prev + (size - bsize)/2;
639 
640  for (j = 0; j < mtab->fmode[ftype].sub; j++) {
641  int sub_wtype = ftype == FT_MEDIUM ? 8 : wtype;
642 
643  if (!j && wtype == 4)
644  sub_wtype = 4;
645  else if (j == mtab->fmode[ftype].sub-1 && wtype == 7)
646  sub_wtype = 7;
647 
648  wsize = types_sizes[wtype_to_wsize[sub_wtype]];
649 
650  mdct->imdct_half(mdct, buf1 + bsize*j, in + bsize*j);
651 
652  tctx->fdsp.vector_fmul_window(out2, prev_buf + (bsize-wsize) / 2,
653  buf1 + bsize * j,
654  ff_sine_windows[av_log2(wsize)],
655  wsize / 2);
656  out2 += wsize;
657 
658  memcpy(out2, buf1 + bsize*j + wsize/2, (bsize - wsize/2)*sizeof(float));
659 
660  out2 += ftype == FT_MEDIUM ? (bsize-wsize)/2 : bsize - wsize;
661 
662  prev_buf = buf1 + bsize*j + bsize/2;
663  }
664 
665  tctx->last_block_pos[ch] = (size + first_wsize)/2;
666 }
667 
668 static void imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype,
669  float **out)
670 {
671  const ModeTab *mtab = tctx->mtab;
672  int size1, size2;
673  float *prev_buf = tctx->prev_frame + tctx->last_block_pos[0];
674  int i;
675 
676  for (i = 0; i < tctx->avctx->channels; i++) {
677  imdct_and_window(tctx, ftype, wtype,
678  tctx->spectrum + i*mtab->size,
679  prev_buf + 2*i*mtab->size,
680  i);
681  }
682 
683  if (!out)
684  return;
685 
686  size2 = tctx->last_block_pos[0];
687  size1 = mtab->size - size2;
688 
689  memcpy(&out[0][0 ], prev_buf, size1 * sizeof(out[0][0]));
690  memcpy(&out[0][size1], tctx->curr_frame, size2 * sizeof(out[0][0]));
691 
692  if (tctx->avctx->channels == 2) {
693  memcpy(&out[1][0], &prev_buf[2*mtab->size], size1 * sizeof(out[1][0]));
694  memcpy(&out[1][size1], &tctx->curr_frame[2*mtab->size], size2 * sizeof(out[1][0]));
695  tctx->fdsp.butterflies_float(out[0], out[1], mtab->size);
696  }
697 }
698 
699 static void dec_bark_env(TwinContext *tctx, const uint8_t *in, int use_hist,
700  int ch, float *out, float gain, enum FrameType ftype)
701 {
702  const ModeTab *mtab = tctx->mtab;
703  int i,j;
704  float *hist = tctx->bark_hist[ftype][ch];
705  float val = ((const float []) {0.4, 0.35, 0.28})[ftype];
706  int bark_n_coef = mtab->fmode[ftype].bark_n_coef;
707  int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef;
708  int idx = 0;
709 
710  for (i = 0; i < fw_cb_len; i++)
711  for (j = 0; j < bark_n_coef; j++, idx++) {
712  float tmp2 =
713  mtab->fmode[ftype].bark_cb[fw_cb_len*in[j] + i] * (1./4096);
714  float st = use_hist ?
715  (1. - val) * tmp2 + val*hist[idx] + 1. : tmp2 + 1.;
716 
717  hist[idx] = tmp2;
718  if (st < -1.) st = 1.;
719 
720  memset_float(out, st * gain, mtab->fmode[ftype].bark_tab[idx]);
721  out += mtab->fmode[ftype].bark_tab[idx];
722  }
723 
724 }
725 
727  float *out, enum FrameType ftype)
728 {
729  const ModeTab *mtab = tctx->mtab;
730  int channels = tctx->avctx->channels;
731  int sub = mtab->fmode[ftype].sub;
732  int block_size = mtab->size / sub;
733  float gain[CHANNELS_MAX*SUBBLOCKS_MAX];
734  float ppc_shape[PPC_SHAPE_LEN_MAX * CHANNELS_MAX * 4];
736  uint8_t bark_use_hist[CHANNELS_MAX][SUBBLOCKS_MAX];
737 
738  uint8_t lpc_idx1[CHANNELS_MAX];
740  uint8_t lpc_hist_idx[CHANNELS_MAX];
741 
742  int i, j, k;
743 
744  dequant(tctx, gb, out, ftype,
745  mtab->fmode[ftype].cb0, mtab->fmode[ftype].cb1,
746  mtab->fmode[ftype].cb_len_read);
747 
748  for (i = 0; i < channels; i++)
749  for (j = 0; j < sub; j++)
750  for (k = 0; k < mtab->fmode[ftype].bark_n_coef; k++)
751  bark1[i][j][k] =
752  get_bits(gb, mtab->fmode[ftype].bark_n_bit);
753 
754  for (i = 0; i < channels; i++)
755  for (j = 0; j < sub; j++)
756  bark_use_hist[i][j] = get_bits1(gb);
757 
758  dec_gain(tctx, gb, ftype, gain);
759 
760  for (i = 0; i < channels; i++) {
761  lpc_hist_idx[i] = get_bits(gb, tctx->mtab->lsp_bit0);
762  lpc_idx1 [i] = get_bits(gb, tctx->mtab->lsp_bit1);
763 
764  for (j = 0; j < tctx->mtab->lsp_split; j++)
765  lpc_idx2[i][j] = get_bits(gb, tctx->mtab->lsp_bit2);
766  }
767 
768  if (ftype == FT_LONG) {
769  int cb_len_p = (tctx->n_div[3] + mtab->ppc_shape_len*channels - 1)/
770  tctx->n_div[3];
771  dequant(tctx, gb, ppc_shape, FT_PPC, mtab->ppc_shape_cb,
772  mtab->ppc_shape_cb + cb_len_p*PPC_SHAPE_CB_SIZE, cb_len_p);
773  }
774 
775  for (i = 0; i < channels; i++) {
776  float *chunk = out + mtab->size * i;
777  float lsp[LSP_COEFS_MAX];
778 
779  for (j = 0; j < sub; j++) {
780  dec_bark_env(tctx, bark1[i][j], bark_use_hist[i][j], i,
781  tctx->tmp_buf, gain[sub*i+j], ftype);
782 
783  tctx->fdsp.vector_fmul(chunk + block_size*j, chunk + block_size*j,
784  tctx->tmp_buf, block_size);
785 
786  }
787 
788  if (ftype == FT_LONG) {
789  float pgain_step = 25000. / ((1 << mtab->pgain_bit) - 1);
790  int p_coef = get_bits(gb, tctx->mtab->ppc_period_bit);
791  int g_coef = get_bits(gb, tctx->mtab->pgain_bit);
792  float v = 1./8192*
793  mulawinv(pgain_step*g_coef+ pgain_step/2, 25000., PGAIN_MU);
794 
795  decode_ppc(tctx, p_coef, ppc_shape + i*mtab->ppc_shape_len, v,
796  chunk);
797  }
798 
799  decode_lsp(tctx, lpc_idx1[i], lpc_idx2[i], lpc_hist_idx[i], lsp,
800  tctx->lsp_hist[i]);
801 
802  dec_lpc_spectrum_inv(tctx, lsp, ftype, tctx->tmp_buf);
803 
804  for (j = 0; j < mtab->fmode[ftype].sub; j++) {
805  tctx->fdsp.vector_fmul(chunk, chunk, tctx->tmp_buf, block_size);
806  chunk += block_size;
807  }
808  }
809 }
810 
811 static int twin_decode_frame(AVCodecContext * avctx, void *data,
812  int *got_frame_ptr, AVPacket *avpkt)
813 {
814  const uint8_t *buf = avpkt->data;
815  int buf_size = avpkt->size;
816  TwinContext *tctx = avctx->priv_data;
817  GetBitContext gb;
818  const ModeTab *mtab = tctx->mtab;
819  float **out = NULL;
820  enum FrameType ftype;
821  int window_type, ret;
822  static const enum FrameType wtype_to_ftype_table[] = {
824  FT_MEDIUM, FT_LONG, FT_LONG, FT_MEDIUM, FT_MEDIUM
825  };
826 
827  if (buf_size*8 < avctx->bit_rate*mtab->size/avctx->sample_rate + 8) {
828  av_log(avctx, AV_LOG_ERROR,
829  "Frame too small (%d bytes). Truncated file?\n", buf_size);
830  return AVERROR(EINVAL);
831  }
832 
833  /* get output buffer */
834  if (tctx->discarded_packets >= 2) {
835  tctx->frame.nb_samples = mtab->size;
836  if ((ret = ff_get_buffer(avctx, &tctx->frame)) < 0) {
837  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
838  return ret;
839  }
840  out = (float **)tctx->frame.extended_data;
841  }
842 
843  init_get_bits(&gb, buf, buf_size * 8);
844  skip_bits(&gb, get_bits(&gb, 8));
845  window_type = get_bits(&gb, WINDOW_TYPE_BITS);
846 
847  if (window_type > 8) {
848  av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
849  return -1;
850  }
851 
852  ftype = wtype_to_ftype_table[window_type];
853 
854  read_and_decode_spectrum(tctx, &gb, tctx->spectrum, ftype);
855 
856  imdct_output(tctx, ftype, window_type, out);
857 
858  FFSWAP(float*, tctx->curr_frame, tctx->prev_frame);
859 
860  if (tctx->discarded_packets < 2) {
861  tctx->discarded_packets++;
862  *got_frame_ptr = 0;
863  return buf_size;
864  }
865 
866  *got_frame_ptr = 1;
867  *(AVFrame *)data = tctx->frame;;
868 
869  return buf_size;
870 }
871 
876 {
877  int i, j, ret;
878  const ModeTab *mtab = tctx->mtab;
879  int size_s = mtab->size / mtab->fmode[FT_SHORT].sub;
880  int size_m = mtab->size / mtab->fmode[FT_MEDIUM].sub;
881  int channels = tctx->avctx->channels;
882  float norm = channels == 1 ? 2. : 1.;
883 
884  for (i = 0; i < 3; i++) {
885  int bsize = tctx->mtab->size/tctx->mtab->fmode[i].sub;
886  if ((ret = ff_mdct_init(&tctx->mdct_ctx[i], av_log2(bsize) + 1, 1,
887  -sqrt(norm/bsize) / (1<<15))))
888  return ret;
889  }
890 
891  FF_ALLOC_OR_GOTO(tctx->avctx, tctx->tmp_buf,
892  mtab->size * sizeof(*tctx->tmp_buf), alloc_fail);
893 
894  FF_ALLOC_OR_GOTO(tctx->avctx, tctx->spectrum,
895  2 * mtab->size * channels * sizeof(*tctx->spectrum),
896  alloc_fail);
897  FF_ALLOC_OR_GOTO(tctx->avctx, tctx->curr_frame,
898  2 * mtab->size * channels * sizeof(*tctx->curr_frame),
899  alloc_fail);
900  FF_ALLOC_OR_GOTO(tctx->avctx, tctx->prev_frame,
901  2 * mtab->size * channels * sizeof(*tctx->prev_frame),
902  alloc_fail);
903 
904  for (i = 0; i < 3; i++) {
905  int m = 4*mtab->size/mtab->fmode[i].sub;
906  double freq = 2*M_PI/m;
907  FF_ALLOC_OR_GOTO(tctx->avctx, tctx->cos_tabs[i],
908  (m / 4) * sizeof(*tctx->cos_tabs[i]), alloc_fail);
909 
910  for (j = 0; j <= m/8; j++)
911  tctx->cos_tabs[i][j] = cos((2*j + 1)*freq);
912  for (j = 1; j < m/8; j++)
913  tctx->cos_tabs[i][m/4-j] = tctx->cos_tabs[i][j];
914  }
915 
916 
918  ff_init_ff_sine_windows(av_log2(size_s/2));
920 
921  return 0;
922 alloc_fail:
923  return AVERROR(ENOMEM);
924 }
925 
932 static void permutate_in_line(int16_t *tab, int num_vect, int num_blocks,
933  int block_size,
934  const uint8_t line_len[2], int length_div,
935  enum FrameType ftype)
936 
937 {
938  int i,j;
939 
940  for (i = 0; i < line_len[0]; i++) {
941  int shift;
942 
943  if (num_blocks == 1 ||
944  (ftype == FT_LONG && num_vect % num_blocks) ||
945  (ftype != FT_LONG && num_vect & 1 ) ||
946  i == line_len[1]) {
947  shift = 0;
948  } else if (ftype == FT_LONG) {
949  shift = i;
950  } else
951  shift = i*i;
952 
953  for (j = 0; j < num_vect && (j+num_vect*i < block_size*num_blocks); j++)
954  tab[i*num_vect+j] = i*num_vect + (j + shift) % num_vect;
955  }
956 }
957 
973 static void transpose_perm(int16_t *out, int16_t *in, int num_vect,
974  const uint8_t line_len[2], int length_div)
975 {
976  int i,j;
977  int cont= 0;
978  for (i = 0; i < num_vect; i++)
979  for (j = 0; j < line_len[i >= length_div]; j++)
980  out[cont++] = in[j*num_vect + i];
981 }
982 
983 static void linear_perm(int16_t *out, int16_t *in, int n_blocks, int size)
984 {
985  int block_size = size/n_blocks;
986  int i;
987 
988  for (i = 0; i < size; i++)
989  out[i] = block_size * (in[i] % n_blocks) + in[i] / n_blocks;
990 }
991 
992 static av_cold void construct_perm_table(TwinContext *tctx,enum FrameType ftype)
993 {
994  int block_size;
995  const ModeTab *mtab = tctx->mtab;
996  int size;
997  int16_t *tmp_perm = (int16_t *) tctx->tmp_buf;
998 
999  if (ftype == FT_PPC) {
1000  size = tctx->avctx->channels;
1001  block_size = mtab->ppc_shape_len;
1002  } else {
1003  size = tctx->avctx->channels * mtab->fmode[ftype].sub;
1004  block_size = mtab->size / mtab->fmode[ftype].sub;
1005  }
1006 
1007  permutate_in_line(tmp_perm, tctx->n_div[ftype], size,
1008  block_size, tctx->length[ftype],
1009  tctx->length_change[ftype], ftype);
1010 
1011  transpose_perm(tctx->permut[ftype], tmp_perm, tctx->n_div[ftype],
1012  tctx->length[ftype], tctx->length_change[ftype]);
1013 
1014  linear_perm(tctx->permut[ftype], tctx->permut[ftype], size,
1015  size*block_size);
1016 }
1017 
1019 {
1020  const ModeTab *mtab = tctx->mtab;
1021  int n_ch = tctx->avctx->channels;
1022  int total_fr_bits = tctx->avctx->bit_rate*mtab->size/
1023  tctx->avctx->sample_rate;
1024 
1025  int lsp_bits_per_block = n_ch*(mtab->lsp_bit0 + mtab->lsp_bit1 +
1026  mtab->lsp_split*mtab->lsp_bit2);
1027 
1028  int ppc_bits = n_ch*(mtab->pgain_bit + mtab->ppc_shape_bit +
1029  mtab->ppc_period_bit);
1030 
1031  int bsize_no_main_cb[3];
1032  int bse_bits[3];
1033  int i;
1034  enum FrameType frametype;
1035 
1036  for (i = 0; i < 3; i++)
1037  // +1 for history usage switch
1038  bse_bits[i] = n_ch *
1039  (mtab->fmode[i].bark_n_coef * mtab->fmode[i].bark_n_bit + 1);
1040 
1041  bsize_no_main_cb[2] = bse_bits[2] + lsp_bits_per_block + ppc_bits +
1042  WINDOW_TYPE_BITS + n_ch*GAIN_BITS;
1043 
1044  for (i = 0; i < 2; i++)
1045  bsize_no_main_cb[i] =
1046  lsp_bits_per_block + n_ch*GAIN_BITS + WINDOW_TYPE_BITS +
1047  mtab->fmode[i].sub*(bse_bits[i] + n_ch*SUB_GAIN_BITS);
1048 
1049  // The remaining bits are all used for the main spectrum coefficients
1050  for (i = 0; i < 4; i++) {
1051  int bit_size;
1052  int vect_size;
1053  int rounded_up, rounded_down, num_rounded_down, num_rounded_up;
1054  if (i == 3) {
1055  bit_size = n_ch * mtab->ppc_shape_bit;
1056  vect_size = n_ch * mtab->ppc_shape_len;
1057  } else {
1058  bit_size = total_fr_bits - bsize_no_main_cb[i];
1059  vect_size = n_ch * mtab->size;
1060  }
1061 
1062  tctx->n_div[i] = (bit_size + 13) / 14;
1063 
1064  rounded_up = (bit_size + tctx->n_div[i] - 1)/tctx->n_div[i];
1065  rounded_down = (bit_size )/tctx->n_div[i];
1066  num_rounded_down = rounded_up * tctx->n_div[i] - bit_size;
1067  num_rounded_up = tctx->n_div[i] - num_rounded_down;
1068  tctx->bits_main_spec[0][i][0] = (rounded_up + 1)/2;
1069  tctx->bits_main_spec[1][i][0] = (rounded_up )/2;
1070  tctx->bits_main_spec[0][i][1] = (rounded_down + 1)/2;
1071  tctx->bits_main_spec[1][i][1] = (rounded_down )/2;
1072  tctx->bits_main_spec_change[i] = num_rounded_up;
1073 
1074  rounded_up = (vect_size + tctx->n_div[i] - 1)/tctx->n_div[i];
1075  rounded_down = (vect_size )/tctx->n_div[i];
1076  num_rounded_down = rounded_up * tctx->n_div[i] - vect_size;
1077  num_rounded_up = tctx->n_div[i] - num_rounded_down;
1078  tctx->length[i][0] = rounded_up;
1079  tctx->length[i][1] = rounded_down;
1080  tctx->length_change[i] = num_rounded_up;
1081  }
1082 
1083  for (frametype = FT_SHORT; frametype <= FT_PPC; frametype++)
1084  construct_perm_table(tctx, frametype);
1085 }
1086 
1088 {
1089  TwinContext *tctx = avctx->priv_data;
1090  int i;
1091 
1092  for (i = 0; i < 3; i++) {
1093  ff_mdct_end(&tctx->mdct_ctx[i]);
1094  av_free(tctx->cos_tabs[i]);
1095  }
1096 
1097 
1098  av_free(tctx->curr_frame);
1099  av_free(tctx->spectrum);
1100  av_free(tctx->prev_frame);
1101  av_free(tctx->tmp_buf);
1102 
1103  return 0;
1104 }
1105 
1107 {
1108  int ret;
1109  TwinContext *tctx = avctx->priv_data;
1110  int isampf, ibps;
1111 
1112  tctx->avctx = avctx;
1113  avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1114 
1115  if (!avctx->extradata || avctx->extradata_size < 12) {
1116  av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n");
1117  return AVERROR_INVALIDDATA;
1118  }
1119  avctx->channels = AV_RB32(avctx->extradata ) + 1;
1120  avctx->bit_rate = AV_RB32(avctx->extradata + 4) * 1000;
1121  isampf = AV_RB32(avctx->extradata + 8);
1122 
1123  if (isampf < 8 || isampf > 44) {
1124  av_log(avctx, AV_LOG_ERROR, "Unsupported sample rate\n");
1125  return AVERROR_INVALIDDATA;
1126  }
1127  switch (isampf) {
1128  case 44: avctx->sample_rate = 44100; break;
1129  case 22: avctx->sample_rate = 22050; break;
1130  case 11: avctx->sample_rate = 11025; break;
1131  default: avctx->sample_rate = isampf * 1000; break;
1132  }
1133 
1134  if (avctx->channels <= 0 || avctx->channels > CHANNELS_MAX) {
1135  av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n",
1136  avctx->channels);
1137  return -1;
1138  }
1139  avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO :
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 
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 = AV_CODEC_ID_TWINVQ,
1183  .priv_data_size = sizeof(TwinContext),
1187  .capabilities = CODEC_CAP_DR1,
1188  .long_name = NULL_IF_CONFIG_SMALL("VQF TwinVQ"),
1189  .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1191 };