FFmpeg  4.1.11
aacpsy.c
Go to the documentation of this file.
1 /*
2  * AAC encoder psychoacoustic model
3  * Copyright (C) 2008 Konstantin Shishkov
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 /**
23  * @file
24  * AAC encoder psychoacoustic model
25  */
26 
27 #include "libavutil/attributes.h"
28 #include "libavutil/ffmath.h"
29 
30 #include "avcodec.h"
31 #include "aactab.h"
32 #include "psymodel.h"
33 
34 /***********************************
35  * TODOs:
36  * try other bitrate controlling mechanism (maybe use ratecontrol.c?)
37  * control quality for quality-based output
38  **********************************/
39 
40 /**
41  * constants for 3GPP AAC psychoacoustic model
42  * @{
43  */
44 #define PSY_3GPP_THR_SPREAD_HI 1.5f // spreading factor for low-to-hi threshold spreading (15 dB/Bark)
45 #define PSY_3GPP_THR_SPREAD_LOW 3.0f // spreading factor for hi-to-low threshold spreading (30 dB/Bark)
46 /* spreading factor for low-to-hi energy spreading, long block, > 22kbps/channel (20dB/Bark) */
47 #define PSY_3GPP_EN_SPREAD_HI_L1 2.0f
48 /* spreading factor for low-to-hi energy spreading, long block, <= 22kbps/channel (15dB/Bark) */
49 #define PSY_3GPP_EN_SPREAD_HI_L2 1.5f
50 /* spreading factor for low-to-hi energy spreading, short block (15 dB/Bark) */
51 #define PSY_3GPP_EN_SPREAD_HI_S 1.5f
52 /* spreading factor for hi-to-low energy spreading, long block (30dB/Bark) */
53 #define PSY_3GPP_EN_SPREAD_LOW_L 3.0f
54 /* spreading factor for hi-to-low energy spreading, short block (20dB/Bark) */
55 #define PSY_3GPP_EN_SPREAD_LOW_S 2.0f
56 
57 #define PSY_3GPP_RPEMIN 0.01f
58 #define PSY_3GPP_RPELEV 2.0f
59 
60 #define PSY_3GPP_C1 3.0f /* log2(8) */
61 #define PSY_3GPP_C2 1.3219281f /* log2(2.5) */
62 #define PSY_3GPP_C3 0.55935729f /* 1 - C2 / C1 */
63 
64 #define PSY_SNR_1DB 7.9432821e-1f /* -1dB */
65 #define PSY_SNR_25DB 3.1622776e-3f /* -25dB */
66 
67 #define PSY_3GPP_SAVE_SLOPE_L -0.46666667f
68 #define PSY_3GPP_SAVE_SLOPE_S -0.36363637f
69 #define PSY_3GPP_SAVE_ADD_L -0.84285712f
70 #define PSY_3GPP_SAVE_ADD_S -0.75f
71 #define PSY_3GPP_SPEND_SLOPE_L 0.66666669f
72 #define PSY_3GPP_SPEND_SLOPE_S 0.81818181f
73 #define PSY_3GPP_SPEND_ADD_L -0.35f
74 #define PSY_3GPP_SPEND_ADD_S -0.26111111f
75 #define PSY_3GPP_CLIP_LO_L 0.2f
76 #define PSY_3GPP_CLIP_LO_S 0.2f
77 #define PSY_3GPP_CLIP_HI_L 0.95f
78 #define PSY_3GPP_CLIP_HI_S 0.75f
79 
80 #define PSY_3GPP_AH_THR_LONG 0.5f
81 #define PSY_3GPP_AH_THR_SHORT 0.63f
82 
83 #define PSY_PE_FORGET_SLOPE 511
84 
85 enum {
89 };
90 
91 #define PSY_3GPP_BITS_TO_PE(bits) ((bits) * 1.18f)
92 #define PSY_3GPP_PE_TO_BITS(bits) ((bits) / 1.18f)
93 
94 /* LAME psy model constants */
95 #define PSY_LAME_FIR_LEN 21 ///< LAME psy model FIR order
96 #define AAC_BLOCK_SIZE_LONG 1024 ///< long block size
97 #define AAC_BLOCK_SIZE_SHORT 128 ///< short block size
98 #define AAC_NUM_BLOCKS_SHORT 8 ///< number of blocks in a short sequence
99 #define PSY_LAME_NUM_SUBBLOCKS 3 ///< Number of sub-blocks in each short block
100 
101 /**
102  * @}
103  */
104 
105 /**
106  * information for single band used by 3GPP TS26.403-inspired psychoacoustic model
107  */
108 typedef struct AacPsyBand{
109  float energy; ///< band energy
110  float thr; ///< energy threshold
111  float thr_quiet; ///< threshold in quiet
112  float nz_lines; ///< number of non-zero spectral lines
113  float active_lines; ///< number of active spectral lines
114  float pe; ///< perceptual entropy
115  float pe_const; ///< constant part of the PE calculation
116  float norm_fac; ///< normalization factor for linearization
117  int avoid_holes; ///< hole avoidance flag
118 }AacPsyBand;
119 
120 /**
121  * single/pair channel context for psychoacoustic model
122  */
123 typedef struct AacPsyChannel{
124  AacPsyBand band[128]; ///< bands information
125  AacPsyBand prev_band[128]; ///< bands information from the previous frame
126 
127  float win_energy; ///< sliding average of channel energy
128  float iir_state[2]; ///< hi-pass IIR filter state
129  uint8_t next_grouping; ///< stored grouping scheme for the next frame (in case of 8 short window sequence)
130  enum WindowSequence next_window_seq; ///< window sequence to be used in the next frame
131  /* LAME psy model specific members */
132  float attack_threshold; ///< attack threshold for this channel
133  float prev_energy_subshort[AAC_NUM_BLOCKS_SHORT * PSY_LAME_NUM_SUBBLOCKS];
134  int prev_attack; ///< attack value for the last short block in the previous sequence
136 
137 /**
138  * psychoacoustic model frame type-dependent coefficients
139  */
140 typedef struct AacPsyCoeffs{
141  float ath; ///< absolute threshold of hearing per bands
142  float barks; ///< Bark value for each spectral band in long frame
143  float spread_low[2]; ///< spreading factor for low-to-high threshold spreading in long frame
144  float spread_hi [2]; ///< spreading factor for high-to-low threshold spreading in long frame
145  float min_snr; ///< minimal SNR
146 }AacPsyCoeffs;
147 
148 /**
149  * 3GPP TS26.403-inspired psychoacoustic model specific data
150  */
151 typedef struct AacPsyContext{
152  int chan_bitrate; ///< bitrate per channel
153  int frame_bits; ///< average bits per frame
154  int fill_level; ///< bit reservoir fill level
155  struct {
156  float min; ///< minimum allowed PE for bit factor calculation
157  float max; ///< maximum allowed PE for bit factor calculation
158  float previous; ///< allowed PE of the previous frame
159  float correction; ///< PE correction factor
160  } pe;
161  AacPsyCoeffs psy_coef[2][64];
163  float global_quality; ///< normalized global quality taken from avctx
165 
166 /**
167  * LAME psy model preset struct
168  */
169 typedef struct PsyLamePreset {
170  int quality; ///< Quality to map the rest of the vaules to.
171  /* This is overloaded to be both kbps per channel in ABR mode, and
172  * requested quality in constant quality mode.
173  */
174  float st_lrm; ///< short threshold for L, R, and M channels
175 } PsyLamePreset;
176 
177 /**
178  * LAME psy model preset table for ABR
179  */
180 static const PsyLamePreset psy_abr_map[] = {
181 /* TODO: Tuning. These were taken from LAME. */
182 /* kbps/ch st_lrm */
183  { 8, 6.60},
184  { 16, 6.60},
185  { 24, 6.60},
186  { 32, 6.60},
187  { 40, 6.60},
188  { 48, 6.60},
189  { 56, 6.60},
190  { 64, 6.40},
191  { 80, 6.00},
192  { 96, 5.60},
193  {112, 5.20},
194  {128, 5.20},
195  {160, 5.20}
196 };
197 
198 /**
199 * LAME psy model preset table for constant quality
200 */
201 static const PsyLamePreset psy_vbr_map[] = {
202 /* vbr_q st_lrm */
203  { 0, 4.20},
204  { 1, 4.20},
205  { 2, 4.20},
206  { 3, 4.20},
207  { 4, 4.20},
208  { 5, 4.20},
209  { 6, 4.20},
210  { 7, 4.20},
211  { 8, 4.20},
212  { 9, 4.20},
213  {10, 4.20}
214 };
215 
216 /**
217  * LAME psy model FIR coefficient table
218  */
219 static const float psy_fir_coeffs[] = {
220  -8.65163e-18 * 2, -0.00851586 * 2, -6.74764e-18 * 2, 0.0209036 * 2,
221  -3.36639e-17 * 2, -0.0438162 * 2, -1.54175e-17 * 2, 0.0931738 * 2,
222  -5.52212e-17 * 2, -0.313819 * 2
223 };
224 
225 #if ARCH_MIPS
226 # include "mips/aacpsy_mips.h"
227 #endif /* ARCH_MIPS */
228 
229 /**
230  * Calculate the ABR attack threshold from the above LAME psymodel table.
231  */
233 {
234  /* Assume max bitrate to start with */
235  int lower_range = 12, upper_range = 12;
236  int lower_range_kbps = psy_abr_map[12].quality;
237  int upper_range_kbps = psy_abr_map[12].quality;
238  int i;
239 
240  /* Determine which bitrates the value specified falls between.
241  * If the loop ends without breaking our above assumption of 320kbps was correct.
242  */
243  for (i = 1; i < 13; i++) {
244  if (FFMAX(bitrate, psy_abr_map[i].quality) != bitrate) {
245  upper_range = i;
246  upper_range_kbps = psy_abr_map[i ].quality;
247  lower_range = i - 1;
248  lower_range_kbps = psy_abr_map[i - 1].quality;
249  break; /* Upper range found */
250  }
251  }
252 
253  /* Determine which range the value specified is closer to */
254  if ((upper_range_kbps - bitrate) > (bitrate - lower_range_kbps))
255  return psy_abr_map[lower_range].st_lrm;
256  return psy_abr_map[upper_range].st_lrm;
257 }
258 
259 /**
260  * LAME psy model specific initialization
261  */
263 {
264  int i, j;
265 
266  for (i = 0; i < avctx->channels; i++) {
267  AacPsyChannel *pch = &ctx->ch[i];
268 
269  if (avctx->flags & AV_CODEC_FLAG_QSCALE)
270  pch->attack_threshold = psy_vbr_map[avctx->global_quality / FF_QP2LAMBDA].st_lrm;
271  else
272  pch->attack_threshold = lame_calc_attack_threshold(avctx->bit_rate / avctx->channels / 1000);
273 
274  for (j = 0; j < AAC_NUM_BLOCKS_SHORT * PSY_LAME_NUM_SUBBLOCKS; j++)
275  pch->prev_energy_subshort[j] = 10.0f;
276  }
277 }
278 
279 /**
280  * Calculate Bark value for given line.
281  */
282 static av_cold float calc_bark(float f)
283 {
284  return 13.3f * atanf(0.00076f * f) + 3.5f * atanf((f / 7500.0f) * (f / 7500.0f));
285 }
286 
287 #define ATH_ADD 4
288 /**
289  * Calculate ATH value for given frequency.
290  * Borrowed from Lame.
291  */
292 static av_cold float ath(float f, float add)
293 {
294  f /= 1000.0f;
295  return 3.64 * pow(f, -0.8)
296  - 6.8 * exp(-0.6 * (f - 3.4) * (f - 3.4))
297  + 6.0 * exp(-0.15 * (f - 8.7) * (f - 8.7))
298  + (0.6 + 0.04 * add) * 0.001 * f * f * f * f;
299 }
300 
302  AacPsyContext *pctx;
303  float bark;
304  int i, j, g, start;
305  float prev, minscale, minath, minsnr, pe_min;
306  int chan_bitrate = ctx->avctx->bit_rate / ((ctx->avctx->flags & AV_CODEC_FLAG_QSCALE) ? 2.0f : ctx->avctx->channels);
307 
308  const int bandwidth = ctx->cutoff ? ctx->cutoff : AAC_CUTOFF(ctx->avctx);
309  const float num_bark = calc_bark((float)bandwidth);
310 
311  if (bandwidth <= 0)
312  return AVERROR(EINVAL);
313 
314  ctx->model_priv_data = av_mallocz(sizeof(AacPsyContext));
315  if (!ctx->model_priv_data)
316  return AVERROR(ENOMEM);
317  pctx = ctx->model_priv_data;
318  pctx->global_quality = (ctx->avctx->global_quality ? ctx->avctx->global_quality : 120) * 0.01f;
319 
320  if (ctx->avctx->flags & AV_CODEC_FLAG_QSCALE) {
321  /* Use the target average bitrate to compute spread parameters */
322  chan_bitrate = (int)(chan_bitrate / 120.0 * (ctx->avctx->global_quality ? ctx->avctx->global_quality : 120));
323  }
324 
325  pctx->chan_bitrate = chan_bitrate;
326  pctx->frame_bits = FFMIN(2560, chan_bitrate * AAC_BLOCK_SIZE_LONG / ctx->avctx->sample_rate);
327  pctx->pe.min = 8.0f * AAC_BLOCK_SIZE_LONG * bandwidth / (ctx->avctx->sample_rate * 2.0f);
328  pctx->pe.max = 12.0f * AAC_BLOCK_SIZE_LONG * bandwidth / (ctx->avctx->sample_rate * 2.0f);
329  ctx->bitres.size = 6144 - pctx->frame_bits;
330  ctx->bitres.size -= ctx->bitres.size % 8;
331  pctx->fill_level = ctx->bitres.size;
332  minath = ath(3410 - 0.733 * ATH_ADD, ATH_ADD);
333  for (j = 0; j < 2; j++) {
334  AacPsyCoeffs *coeffs = pctx->psy_coef[j];
335  const uint8_t *band_sizes = ctx->bands[j];
336  float line_to_frequency = ctx->avctx->sample_rate / (j ? 256.f : 2048.0f);
337  float avg_chan_bits = chan_bitrate * (j ? 128.0f : 1024.0f) / ctx->avctx->sample_rate;
338  /* reference encoder uses 2.4% here instead of 60% like the spec says */
339  float bark_pe = 0.024f * PSY_3GPP_BITS_TO_PE(avg_chan_bits) / num_bark;
340  float en_spread_low = j ? PSY_3GPP_EN_SPREAD_LOW_S : PSY_3GPP_EN_SPREAD_LOW_L;
341  /* High energy spreading for long blocks <= 22kbps/channel and short blocks are the same. */
342  float en_spread_hi = (j || (chan_bitrate <= 22.0f)) ? PSY_3GPP_EN_SPREAD_HI_S : PSY_3GPP_EN_SPREAD_HI_L1;
343 
344  i = 0;
345  prev = 0.0;
346  for (g = 0; g < ctx->num_bands[j]; g++) {
347  i += band_sizes[g];
348  bark = calc_bark((i-1) * line_to_frequency);
349  coeffs[g].barks = (bark + prev) / 2.0;
350  prev = bark;
351  }
352  for (g = 0; g < ctx->num_bands[j] - 1; g++) {
353  AacPsyCoeffs *coeff = &coeffs[g];
354  float bark_width = coeffs[g+1].barks - coeffs->barks;
355  coeff->spread_low[0] = ff_exp10(-bark_width * PSY_3GPP_THR_SPREAD_LOW);
356  coeff->spread_hi [0] = ff_exp10(-bark_width * PSY_3GPP_THR_SPREAD_HI);
357  coeff->spread_low[1] = ff_exp10(-bark_width * en_spread_low);
358  coeff->spread_hi [1] = ff_exp10(-bark_width * en_spread_hi);
359  pe_min = bark_pe * bark_width;
360  minsnr = exp2(pe_min / band_sizes[g]) - 1.5f;
361  coeff->min_snr = av_clipf(1.0f / minsnr, PSY_SNR_25DB, PSY_SNR_1DB);
362  }
363  start = 0;
364  for (g = 0; g < ctx->num_bands[j]; g++) {
365  minscale = ath(start * line_to_frequency, ATH_ADD);
366  for (i = 1; i < band_sizes[g]; i++)
367  minscale = FFMIN(minscale, ath((start + i) * line_to_frequency, ATH_ADD));
368  coeffs[g].ath = minscale - minath;
369  start += band_sizes[g];
370  }
371  }
372 
373  pctx->ch = av_mallocz_array(ctx->avctx->channels, sizeof(AacPsyChannel));
374  if (!pctx->ch) {
375  av_freep(&ctx->model_priv_data);
376  return AVERROR(ENOMEM);
377  }
378 
379  lame_window_init(pctx, ctx->avctx);
380 
381  return 0;
382 }
383 
384 /**
385  * IIR filter used in block switching decision
386  */
387 static float iir_filter(int in, float state[2])
388 {
389  float ret;
390 
391  ret = 0.7548f * (in - state[0]) + 0.5095f * state[1];
392  state[0] = in;
393  state[1] = ret;
394  return ret;
395 }
396 
397 /**
398  * window grouping information stored as bits (0 - new group, 1 - group continues)
399  */
400 static const uint8_t window_grouping[9] = {
401  0xB6, 0x6C, 0xD8, 0xB2, 0x66, 0xC6, 0x96, 0x36, 0x36
402 };
403 
404 /**
405  * Tell encoder which window types to use.
406  * @see 3GPP TS26.403 5.4.1 "Blockswitching"
407  */
409  const int16_t *audio,
410  const int16_t *la,
411  int channel, int prev_type)
412 {
413  int i, j;
414  int br = ((AacPsyContext*)ctx->model_priv_data)->chan_bitrate;
415  int attack_ratio = br <= 16000 ? 18 : 10;
417  AacPsyChannel *pch = &pctx->ch[channel];
418  uint8_t grouping = 0;
419  int next_type = pch->next_window_seq;
420  FFPsyWindowInfo wi = { { 0 } };
421 
422  if (la) {
423  float s[8], v;
424  int switch_to_eight = 0;
425  float sum = 0.0, sum2 = 0.0;
426  int attack_n = 0;
427  int stay_short = 0;
428  for (i = 0; i < 8; i++) {
429  for (j = 0; j < 128; j++) {
430  v = iir_filter(la[i*128+j], pch->iir_state);
431  sum += v*v;
432  }
433  s[i] = sum;
434  sum2 += sum;
435  }
436  for (i = 0; i < 8; i++) {
437  if (s[i] > pch->win_energy * attack_ratio) {
438  attack_n = i + 1;
439  switch_to_eight = 1;
440  break;
441  }
442  }
443  pch->win_energy = pch->win_energy*7/8 + sum2/64;
444 
445  wi.window_type[1] = prev_type;
446  switch (prev_type) {
447  case ONLY_LONG_SEQUENCE:
448  wi.window_type[0] = switch_to_eight ? LONG_START_SEQUENCE : ONLY_LONG_SEQUENCE;
449  next_type = switch_to_eight ? EIGHT_SHORT_SEQUENCE : ONLY_LONG_SEQUENCE;
450  break;
451  case LONG_START_SEQUENCE:
452  wi.window_type[0] = EIGHT_SHORT_SEQUENCE;
453  grouping = pch->next_grouping;
454  next_type = switch_to_eight ? EIGHT_SHORT_SEQUENCE : LONG_STOP_SEQUENCE;
455  break;
456  case LONG_STOP_SEQUENCE:
457  wi.window_type[0] = switch_to_eight ? LONG_START_SEQUENCE : ONLY_LONG_SEQUENCE;
458  next_type = switch_to_eight ? EIGHT_SHORT_SEQUENCE : ONLY_LONG_SEQUENCE;
459  break;
461  stay_short = next_type == EIGHT_SHORT_SEQUENCE || switch_to_eight;
462  wi.window_type[0] = stay_short ? EIGHT_SHORT_SEQUENCE : LONG_STOP_SEQUENCE;
463  grouping = next_type == EIGHT_SHORT_SEQUENCE ? pch->next_grouping : 0;
464  next_type = switch_to_eight ? EIGHT_SHORT_SEQUENCE : LONG_STOP_SEQUENCE;
465  break;
466  }
467 
468  pch->next_grouping = window_grouping[attack_n];
469  pch->next_window_seq = next_type;
470  } else {
471  for (i = 0; i < 3; i++)
472  wi.window_type[i] = prev_type;
473  grouping = (prev_type == EIGHT_SHORT_SEQUENCE) ? window_grouping[0] : 0;
474  }
475 
476  wi.window_shape = 1;
477  if (wi.window_type[0] != EIGHT_SHORT_SEQUENCE) {
478  wi.num_windows = 1;
479  wi.grouping[0] = 1;
480  } else {
481  int lastgrp = 0;
482  wi.num_windows = 8;
483  for (i = 0; i < 8; i++) {
484  if (!((grouping >> i) & 1))
485  lastgrp = i;
486  wi.grouping[lastgrp]++;
487  }
488  }
489 
490  return wi;
491 }
492 
493 /* 5.6.1.2 "Calculation of Bit Demand" */
494 static int calc_bit_demand(AacPsyContext *ctx, float pe, int bits, int size,
495  int short_window)
496 {
497  const float bitsave_slope = short_window ? PSY_3GPP_SAVE_SLOPE_S : PSY_3GPP_SAVE_SLOPE_L;
498  const float bitsave_add = short_window ? PSY_3GPP_SAVE_ADD_S : PSY_3GPP_SAVE_ADD_L;
499  const float bitspend_slope = short_window ? PSY_3GPP_SPEND_SLOPE_S : PSY_3GPP_SPEND_SLOPE_L;
500  const float bitspend_add = short_window ? PSY_3GPP_SPEND_ADD_S : PSY_3GPP_SPEND_ADD_L;
501  const float clip_low = short_window ? PSY_3GPP_CLIP_LO_S : PSY_3GPP_CLIP_LO_L;
502  const float clip_high = short_window ? PSY_3GPP_CLIP_HI_S : PSY_3GPP_CLIP_HI_L;
503  float clipped_pe, bit_save, bit_spend, bit_factor, fill_level, forgetful_min_pe;
504 
505  ctx->fill_level += ctx->frame_bits - bits;
506  ctx->fill_level = av_clip(ctx->fill_level, 0, size);
507  fill_level = av_clipf((float)ctx->fill_level / size, clip_low, clip_high);
508  clipped_pe = av_clipf(pe, ctx->pe.min, ctx->pe.max);
509  bit_save = (fill_level + bitsave_add) * bitsave_slope;
510  assert(bit_save <= 0.3f && bit_save >= -0.05000001f);
511  bit_spend = (fill_level + bitspend_add) * bitspend_slope;
512  assert(bit_spend <= 0.5f && bit_spend >= -0.1f);
513  /* The bit factor graph in the spec is obviously incorrect.
514  * bit_spend + ((bit_spend - bit_spend))...
515  * The reference encoder subtracts everything from 1, but also seems incorrect.
516  * 1 - bit_save + ((bit_spend + bit_save))...
517  * Hopefully below is correct.
518  */
519  bit_factor = 1.0f - bit_save + ((bit_spend - bit_save) / (ctx->pe.max - ctx->pe.min)) * (clipped_pe - ctx->pe.min);
520  /* NOTE: The reference encoder attempts to center pe max/min around the current pe.
521  * Here we do that by slowly forgetting pe.min when pe stays in a range that makes
522  * it unlikely (ie: above the mean)
523  */
524  ctx->pe.max = FFMAX(pe, ctx->pe.max);
525  forgetful_min_pe = ((ctx->pe.min * PSY_PE_FORGET_SLOPE)
526  + FFMAX(ctx->pe.min, pe * (pe / ctx->pe.max))) / (PSY_PE_FORGET_SLOPE + 1);
527  ctx->pe.min = FFMIN(pe, forgetful_min_pe);
528 
529  /* NOTE: allocate a minimum of 1/8th average frame bits, to avoid
530  * reservoir starvation from producing zero-bit frames
531  */
532  return FFMIN(
533  ctx->frame_bits * bit_factor,
534  FFMAX(ctx->frame_bits + size - bits, ctx->frame_bits / 8));
535 }
536 
537 static float calc_pe_3gpp(AacPsyBand *band)
538 {
539  float pe, a;
540 
541  band->pe = 0.0f;
542  band->pe_const = 0.0f;
543  band->active_lines = 0.0f;
544  if (band->energy > band->thr) {
545  a = log2f(band->energy);
546  pe = a - log2f(band->thr);
547  band->active_lines = band->nz_lines;
548  if (pe < PSY_3GPP_C1) {
549  pe = pe * PSY_3GPP_C3 + PSY_3GPP_C2;
550  a = a * PSY_3GPP_C3 + PSY_3GPP_C2;
551  band->active_lines *= PSY_3GPP_C3;
552  }
553  band->pe = pe * band->nz_lines;
554  band->pe_const = a * band->nz_lines;
555  }
556 
557  return band->pe;
558 }
559 
560 static float calc_reduction_3gpp(float a, float desired_pe, float pe,
561  float active_lines)
562 {
563  float thr_avg, reduction;
564 
565  if(active_lines == 0.0)
566  return 0;
567 
568  thr_avg = exp2f((a - pe) / (4.0f * active_lines));
569  reduction = exp2f((a - desired_pe) / (4.0f * active_lines)) - thr_avg;
570 
571  return FFMAX(reduction, 0.0f);
572 }
573 
574 static float calc_reduced_thr_3gpp(AacPsyBand *band, float min_snr,
575  float reduction)
576 {
577  float thr = band->thr;
578 
579  if (band->energy > thr) {
580  thr = sqrtf(thr);
581  thr = sqrtf(thr) + reduction;
582  thr *= thr;
583  thr *= thr;
584 
585  /* This deviates from the 3GPP spec to match the reference encoder.
586  * It performs min(thr_reduced, max(thr, energy/min_snr)) only for bands
587  * that have hole avoidance on (active or inactive). It always reduces the
588  * threshold of bands with hole avoidance off.
589  */
590  if (thr > band->energy * min_snr && band->avoid_holes != PSY_3GPP_AH_NONE) {
591  thr = FFMAX(band->thr, band->energy * min_snr);
593  }
594  }
595 
596  return thr;
597 }
598 
599 #ifndef calc_thr_3gpp
600 static void calc_thr_3gpp(const FFPsyWindowInfo *wi, const int num_bands, AacPsyChannel *pch,
601  const uint8_t *band_sizes, const float *coefs, const int cutoff)
602 {
603  int i, w, g;
604  int start = 0, wstart = 0;
605  for (w = 0; w < wi->num_windows*16; w += 16) {
606  wstart = 0;
607  for (g = 0; g < num_bands; g++) {
608  AacPsyBand *band = &pch->band[w+g];
609 
610  float form_factor = 0.0f;
611  float Temp;
612  band->energy = 0.0f;
613  if (wstart < cutoff) {
614  for (i = 0; i < band_sizes[g]; i++) {
615  band->energy += coefs[start+i] * coefs[start+i];
616  form_factor += sqrtf(fabs(coefs[start+i]));
617  }
618  }
619  Temp = band->energy > 0 ? sqrtf((float)band_sizes[g] / band->energy) : 0;
620  band->thr = band->energy * 0.001258925f;
621  band->nz_lines = form_factor * sqrtf(Temp);
622 
623  start += band_sizes[g];
624  wstart += band_sizes[g];
625  }
626  }
627 }
628 #endif /* calc_thr_3gpp */
629 
630 #ifndef psy_hp_filter
631 static void psy_hp_filter(const float *firbuf, float *hpfsmpl, const float *psy_fir_coeffs)
632 {
633  int i, j;
634  for (i = 0; i < AAC_BLOCK_SIZE_LONG; i++) {
635  float sum1, sum2;
636  sum1 = firbuf[i + (PSY_LAME_FIR_LEN - 1) / 2];
637  sum2 = 0.0;
638  for (j = 0; j < ((PSY_LAME_FIR_LEN - 1) / 2) - 1; j += 2) {
639  sum1 += psy_fir_coeffs[j] * (firbuf[i + j] + firbuf[i + PSY_LAME_FIR_LEN - j]);
640  sum2 += psy_fir_coeffs[j + 1] * (firbuf[i + j + 1] + firbuf[i + PSY_LAME_FIR_LEN - j - 1]);
641  }
642  /* NOTE: The LAME psymodel expects it's input in the range -32768 to 32768.
643  * Tuning this for normalized floats would be difficult. */
644  hpfsmpl[i] = (sum1 + sum2) * 32768.0f;
645  }
646 }
647 #endif /* psy_hp_filter */
648 
649 /**
650  * Calculate band thresholds as suggested in 3GPP TS26.403
651  */
653  const float *coefs, const FFPsyWindowInfo *wi)
654 {
656  AacPsyChannel *pch = &pctx->ch[channel];
657  int i, w, g;
658  float desired_bits, desired_pe, delta_pe, reduction= NAN, spread_en[128] = {0};
659  float a = 0.0f, active_lines = 0.0f, norm_fac = 0.0f;
660  float pe = pctx->chan_bitrate > 32000 ? 0.0f : FFMAX(50.0f, 100.0f - pctx->chan_bitrate * 100.0f / 32000.0f);
661  const int num_bands = ctx->num_bands[wi->num_windows == 8];
662  const uint8_t *band_sizes = ctx->bands[wi->num_windows == 8];
663  AacPsyCoeffs *coeffs = pctx->psy_coef[wi->num_windows == 8];
664  const float avoid_hole_thr = wi->num_windows == 8 ? PSY_3GPP_AH_THR_SHORT : PSY_3GPP_AH_THR_LONG;
665  const int bandwidth = ctx->cutoff ? ctx->cutoff : AAC_CUTOFF(ctx->avctx);
666  const int cutoff = bandwidth * 2048 / wi->num_windows / ctx->avctx->sample_rate;
667 
668  //calculate energies, initial thresholds and related values - 5.4.2 "Threshold Calculation"
669  calc_thr_3gpp(wi, num_bands, pch, band_sizes, coefs, cutoff);
670 
671  //modify thresholds and energies - spread, threshold in quiet, pre-echo control
672  for (w = 0; w < wi->num_windows*16; w += 16) {
673  AacPsyBand *bands = &pch->band[w];
674 
675  /* 5.4.2.3 "Spreading" & 5.4.3 "Spread Energy Calculation" */
676  spread_en[0] = bands[0].energy;
677  for (g = 1; g < num_bands; g++) {
678  bands[g].thr = FFMAX(bands[g].thr, bands[g-1].thr * coeffs[g].spread_hi[0]);
679  spread_en[w+g] = FFMAX(bands[g].energy, spread_en[w+g-1] * coeffs[g].spread_hi[1]);
680  }
681  for (g = num_bands - 2; g >= 0; g--) {
682  bands[g].thr = FFMAX(bands[g].thr, bands[g+1].thr * coeffs[g].spread_low[0]);
683  spread_en[w+g] = FFMAX(spread_en[w+g], spread_en[w+g+1] * coeffs[g].spread_low[1]);
684  }
685  //5.4.2.4 "Threshold in quiet"
686  for (g = 0; g < num_bands; g++) {
687  AacPsyBand *band = &bands[g];
688 
689  band->thr_quiet = band->thr = FFMAX(band->thr, coeffs[g].ath);
690  //5.4.2.5 "Pre-echo control"
691  if (!(wi->window_type[0] == LONG_STOP_SEQUENCE || (!w && wi->window_type[1] == LONG_START_SEQUENCE)))
692  band->thr = FFMAX(PSY_3GPP_RPEMIN*band->thr, FFMIN(band->thr,
693  PSY_3GPP_RPELEV*pch->prev_band[w+g].thr_quiet));
694 
695  /* 5.6.1.3.1 "Preparatory steps of the perceptual entropy calculation" */
696  pe += calc_pe_3gpp(band);
697  a += band->pe_const;
698  active_lines += band->active_lines;
699 
700  /* 5.6.1.3.3 "Selection of the bands for avoidance of holes" */
701  if (spread_en[w+g] * avoid_hole_thr > band->energy || coeffs[g].min_snr > 1.0f)
703  else
705  }
706  }
707 
708  /* 5.6.1.3.2 "Calculation of the desired perceptual entropy" */
709  ctx->ch[channel].entropy = pe;
710  if (ctx->avctx->flags & AV_CODEC_FLAG_QSCALE) {
711  /* (2.5 * 120) achieves almost transparent rate, and we want to give
712  * ample room downwards, so we make that equivalent to QSCALE=2.4
713  */
714  desired_pe = pe * (ctx->avctx->global_quality ? ctx->avctx->global_quality : 120) / (2 * 2.5f * 120.0f);
715  desired_bits = FFMIN(2560, PSY_3GPP_PE_TO_BITS(desired_pe));
716  desired_pe = PSY_3GPP_BITS_TO_PE(desired_bits); // reflect clipping
717 
718  /* PE slope smoothing */
719  if (ctx->bitres.bits > 0) {
720  desired_bits = FFMIN(2560, PSY_3GPP_PE_TO_BITS(desired_pe));
721  desired_pe = PSY_3GPP_BITS_TO_PE(desired_bits); // reflect clipping
722  }
723 
724  pctx->pe.max = FFMAX(pe, pctx->pe.max);
725  pctx->pe.min = FFMIN(pe, pctx->pe.min);
726  } else {
727  desired_bits = calc_bit_demand(pctx, pe, ctx->bitres.bits, ctx->bitres.size, wi->num_windows == 8);
728  desired_pe = PSY_3GPP_BITS_TO_PE(desired_bits);
729 
730  /* NOTE: PE correction is kept simple. During initial testing it had very
731  * little effect on the final bitrate. Probably a good idea to come
732  * back and do more testing later.
733  */
734  if (ctx->bitres.bits > 0)
735  desired_pe *= av_clipf(pctx->pe.previous / PSY_3GPP_BITS_TO_PE(ctx->bitres.bits),
736  0.85f, 1.15f);
737  }
738  pctx->pe.previous = PSY_3GPP_BITS_TO_PE(desired_bits);
739  ctx->bitres.alloc = desired_bits;
740 
741  if (desired_pe < pe) {
742  /* 5.6.1.3.4 "First Estimation of the reduction value" */
743  for (w = 0; w < wi->num_windows*16; w += 16) {
744  reduction = calc_reduction_3gpp(a, desired_pe, pe, active_lines);
745  pe = 0.0f;
746  a = 0.0f;
747  active_lines = 0.0f;
748  for (g = 0; g < num_bands; g++) {
749  AacPsyBand *band = &pch->band[w+g];
750 
751  band->thr = calc_reduced_thr_3gpp(band, coeffs[g].min_snr, reduction);
752  /* recalculate PE */
753  pe += calc_pe_3gpp(band);
754  a += band->pe_const;
755  active_lines += band->active_lines;
756  }
757  }
758 
759  /* 5.6.1.3.5 "Second Estimation of the reduction value" */
760  for (i = 0; i < 2; i++) {
761  float pe_no_ah = 0.0f, desired_pe_no_ah;
762  active_lines = a = 0.0f;
763  for (w = 0; w < wi->num_windows*16; w += 16) {
764  for (g = 0; g < num_bands; g++) {
765  AacPsyBand *band = &pch->band[w+g];
766 
767  if (band->avoid_holes != PSY_3GPP_AH_ACTIVE) {
768  pe_no_ah += band->pe;
769  a += band->pe_const;
770  active_lines += band->active_lines;
771  }
772  }
773  }
774  desired_pe_no_ah = FFMAX(desired_pe - (pe - pe_no_ah), 0.0f);
775  if (active_lines > 0.0f)
776  reduction = calc_reduction_3gpp(a, desired_pe_no_ah, pe_no_ah, active_lines);
777 
778  pe = 0.0f;
779  for (w = 0; w < wi->num_windows*16; w += 16) {
780  for (g = 0; g < num_bands; g++) {
781  AacPsyBand *band = &pch->band[w+g];
782 
783  if (active_lines > 0.0f)
784  band->thr = calc_reduced_thr_3gpp(band, coeffs[g].min_snr, reduction);
785  pe += calc_pe_3gpp(band);
786  if (band->thr > 0.0f)
787  band->norm_fac = band->active_lines / band->thr;
788  else
789  band->norm_fac = 0.0f;
790  norm_fac += band->norm_fac;
791  }
792  }
793  delta_pe = desired_pe - pe;
794  if (fabs(delta_pe) > 0.05f * desired_pe)
795  break;
796  }
797 
798  if (pe < 1.15f * desired_pe) {
799  /* 6.6.1.3.6 "Final threshold modification by linearization" */
800  norm_fac = norm_fac ? 1.0f / norm_fac : 0;
801  for (w = 0; w < wi->num_windows*16; w += 16) {
802  for (g = 0; g < num_bands; g++) {
803  AacPsyBand *band = &pch->band[w+g];
804 
805  if (band->active_lines > 0.5f) {
806  float delta_sfb_pe = band->norm_fac * norm_fac * delta_pe;
807  float thr = band->thr;
808 
809  thr *= exp2f(delta_sfb_pe / band->active_lines);
810  if (thr > coeffs[g].min_snr * band->energy && band->avoid_holes == PSY_3GPP_AH_INACTIVE)
811  thr = FFMAX(band->thr, coeffs[g].min_snr * band->energy);
812  band->thr = thr;
813  }
814  }
815  }
816  } else {
817  /* 5.6.1.3.7 "Further perceptual entropy reduction" */
818  g = num_bands;
819  while (pe > desired_pe && g--) {
820  for (w = 0; w < wi->num_windows*16; w+= 16) {
821  AacPsyBand *band = &pch->band[w+g];
822  if (band->avoid_holes != PSY_3GPP_AH_NONE && coeffs[g].min_snr < PSY_SNR_1DB) {
823  coeffs[g].min_snr = PSY_SNR_1DB;
824  band->thr = band->energy * PSY_SNR_1DB;
825  pe += band->active_lines * 1.5f - band->pe;
826  }
827  }
828  }
829  /* TODO: allow more holes (unused without mid/side) */
830  }
831  }
832 
833  for (w = 0; w < wi->num_windows*16; w += 16) {
834  for (g = 0; g < num_bands; g++) {
835  AacPsyBand *band = &pch->band[w+g];
836  FFPsyBand *psy_band = &ctx->ch[channel].psy_bands[w+g];
837 
838  psy_band->threshold = band->thr;
839  psy_band->energy = band->energy;
840  psy_band->spread = band->active_lines * 2.0f / band_sizes[g];
841  psy_band->bits = PSY_3GPP_PE_TO_BITS(band->pe);
842  }
843  }
844 
845  memcpy(pch->prev_band, pch->band, sizeof(pch->band));
846 }
847 
849  const float **coeffs, const FFPsyWindowInfo *wi)
850 {
851  int ch;
852  FFPsyChannelGroup *group = ff_psy_find_group(ctx, channel);
853 
854  for (ch = 0; ch < group->num_ch; ch++)
855  psy_3gpp_analyze_channel(ctx, channel + ch, coeffs[ch], &wi[ch]);
856 }
857 
859 {
861  av_freep(&pctx->ch);
862  av_freep(&apc->model_priv_data);
863 }
864 
865 static void lame_apply_block_type(AacPsyChannel *ctx, FFPsyWindowInfo *wi, int uselongblock)
866 {
867  int blocktype = ONLY_LONG_SEQUENCE;
868  if (uselongblock) {
870  blocktype = LONG_STOP_SEQUENCE;
871  } else {
872  blocktype = EIGHT_SHORT_SEQUENCE;
877  }
878 
879  wi->window_type[0] = ctx->next_window_seq;
880  ctx->next_window_seq = blocktype;
881 }
882 
883 static FFPsyWindowInfo psy_lame_window(FFPsyContext *ctx, const float *audio,
884  const float *la, int channel, int prev_type)
885 {
887  AacPsyChannel *pch = &pctx->ch[channel];
888  int grouping = 0;
889  int uselongblock = 1;
890  int attacks[AAC_NUM_BLOCKS_SHORT + 1] = { 0 };
891  int i;
892  FFPsyWindowInfo wi = { { 0 } };
893 
894  if (la) {
895  float hpfsmpl[AAC_BLOCK_SIZE_LONG];
896  const float *pf = hpfsmpl;
897  float attack_intensity[(AAC_NUM_BLOCKS_SHORT + 1) * PSY_LAME_NUM_SUBBLOCKS];
898  float energy_subshort[(AAC_NUM_BLOCKS_SHORT + 1) * PSY_LAME_NUM_SUBBLOCKS];
899  float energy_short[AAC_NUM_BLOCKS_SHORT + 1] = { 0 };
900  const float *firbuf = la + (AAC_BLOCK_SIZE_SHORT/4 - PSY_LAME_FIR_LEN);
901  int att_sum = 0;
902 
903  /* LAME comment: apply high pass filter of fs/4 */
904  psy_hp_filter(firbuf, hpfsmpl, psy_fir_coeffs);
905 
906  /* Calculate the energies of each sub-shortblock */
907  for (i = 0; i < PSY_LAME_NUM_SUBBLOCKS; i++) {
908  energy_subshort[i] = pch->prev_energy_subshort[i + ((AAC_NUM_BLOCKS_SHORT - 1) * PSY_LAME_NUM_SUBBLOCKS)];
909  assert(pch->prev_energy_subshort[i + ((AAC_NUM_BLOCKS_SHORT - 2) * PSY_LAME_NUM_SUBBLOCKS + 1)] > 0);
910  attack_intensity[i] = energy_subshort[i] / pch->prev_energy_subshort[i + ((AAC_NUM_BLOCKS_SHORT - 2) * PSY_LAME_NUM_SUBBLOCKS + 1)];
911  energy_short[0] += energy_subshort[i];
912  }
913 
914  for (i = 0; i < AAC_NUM_BLOCKS_SHORT * PSY_LAME_NUM_SUBBLOCKS; i++) {
915  const float *const pfe = pf + AAC_BLOCK_SIZE_LONG / (AAC_NUM_BLOCKS_SHORT * PSY_LAME_NUM_SUBBLOCKS);
916  float p = 1.0f;
917  for (; pf < pfe; pf++)
918  p = FFMAX(p, fabsf(*pf));
919  pch->prev_energy_subshort[i] = energy_subshort[i + PSY_LAME_NUM_SUBBLOCKS] = p;
920  energy_short[1 + i / PSY_LAME_NUM_SUBBLOCKS] += p;
921  /* NOTE: The indexes below are [i + 3 - 2] in the LAME source.
922  * Obviously the 3 and 2 have some significance, or this would be just [i + 1]
923  * (which is what we use here). What the 3 stands for is ambiguous, as it is both
924  * number of short blocks, and the number of sub-short blocks.
925  * It seems that LAME is comparing each sub-block to sub-block + 1 in the
926  * previous block.
927  */
928  if (p > energy_subshort[i + 1])
929  p = p / energy_subshort[i + 1];
930  else if (energy_subshort[i + 1] > p * 10.0f)
931  p = energy_subshort[i + 1] / (p * 10.0f);
932  else
933  p = 0.0;
934  attack_intensity[i + PSY_LAME_NUM_SUBBLOCKS] = p;
935  }
936 
937  /* compare energy between sub-short blocks */
938  for (i = 0; i < (AAC_NUM_BLOCKS_SHORT + 1) * PSY_LAME_NUM_SUBBLOCKS; i++)
939  if (!attacks[i / PSY_LAME_NUM_SUBBLOCKS])
940  if (attack_intensity[i] > pch->attack_threshold)
941  attacks[i / PSY_LAME_NUM_SUBBLOCKS] = (i % PSY_LAME_NUM_SUBBLOCKS) + 1;
942 
943  /* should have energy change between short blocks, in order to avoid periodic signals */
944  /* Good samples to show the effect are Trumpet test songs */
945  /* GB: tuned (1) to avoid too many short blocks for test sample TRUMPET */
946  /* RH: tuned (2) to let enough short blocks through for test sample FSOL and SNAPS */
947  for (i = 1; i < AAC_NUM_BLOCKS_SHORT + 1; i++) {
948  const float u = energy_short[i - 1];
949  const float v = energy_short[i];
950  const float m = FFMAX(u, v);
951  if (m < 40000) { /* (2) */
952  if (u < 1.7f * v && v < 1.7f * u) { /* (1) */
953  if (i == 1 && attacks[0] < attacks[i])
954  attacks[0] = 0;
955  attacks[i] = 0;
956  }
957  }
958  att_sum += attacks[i];
959  }
960 
961  if (attacks[0] <= pch->prev_attack)
962  attacks[0] = 0;
963 
964  att_sum += attacks[0];
965  /* 3 below indicates the previous attack happened in the last sub-block of the previous sequence */
966  if (pch->prev_attack == 3 || att_sum) {
967  uselongblock = 0;
968 
969  for (i = 1; i < AAC_NUM_BLOCKS_SHORT + 1; i++)
970  if (attacks[i] && attacks[i-1])
971  attacks[i] = 0;
972  }
973  } else {
974  /* We have no lookahead info, so just use same type as the previous sequence. */
975  uselongblock = !(prev_type == EIGHT_SHORT_SEQUENCE);
976  }
977 
978  lame_apply_block_type(pch, &wi, uselongblock);
979 
980  wi.window_type[1] = prev_type;
981  if (wi.window_type[0] != EIGHT_SHORT_SEQUENCE) {
982 
983  wi.num_windows = 1;
984  wi.grouping[0] = 1;
985  if (wi.window_type[0] == LONG_START_SEQUENCE)
986  wi.window_shape = 0;
987  else
988  wi.window_shape = 1;
989 
990  } else {
991  int lastgrp = 0;
992 
993  wi.num_windows = 8;
994  wi.window_shape = 0;
995  for (i = 0; i < 8; i++) {
996  if (!((pch->next_grouping >> i) & 1))
997  lastgrp = i;
998  wi.grouping[lastgrp]++;
999  }
1000  }
1001 
1002  /* Determine grouping, based on the location of the first attack, and save for
1003  * the next frame.
1004  * FIXME: Move this to analysis.
1005  * TODO: Tune groupings depending on attack location
1006  * TODO: Handle more than one attack in a group
1007  */
1008  for (i = 0; i < 9; i++) {
1009  if (attacks[i]) {
1010  grouping = i;
1011  break;
1012  }
1013  }
1014  pch->next_grouping = window_grouping[grouping];
1015 
1016  pch->prev_attack = attacks[8];
1017 
1018  return wi;
1019 }
1020 
1022 {
1023  .name = "3GPP TS 26.403-inspired model",
1024  .init = psy_3gpp_init,
1025  .window = psy_lame_window,
1026  .analyze = psy_3gpp_analyze,
1027  .end = psy_3gpp_end,
1028 };
int quality
Quality to map the rest of the vaules to.
Definition: aacpsy.c:170
float global_quality
normalized global quality taken from avctx
Definition: aacpsy.c:163
int size
static const uint8_t window_grouping[9]
window grouping information stored as bits (0 - new group, 1 - group continues)
Definition: aacpsy.c:400
int grouping[8]
window grouping (for e.g. AAC)
Definition: psymodel.h:81
#define AAC_BLOCK_SIZE_SHORT
short block size
Definition: aacpsy.c:97
static int calc_bit_demand(AacPsyContext *ctx, float pe, int bits, int size, int short_window)
Definition: aacpsy.c:494
uint8_t ** bands
scalefactor band sizes for possible frame sizes
Definition: psymodel.h:98
#define PSY_3GPP_AH_THR_SHORT
Definition: aacpsy.c:81
int64_t bit_rate
the average bitrate
Definition: avcodec.h:1583
const char * g
Definition: vf_curves.c:115
static const PsyLamePreset psy_vbr_map[]
LAME psy model preset table for constant quality.
Definition: aacpsy.c:201
psychoacoustic information for an arbitrary group of channels
Definition: psymodel.h:68
uint8_t pi<< 24) CONV_FUNC(AV_SAMPLE_FMT_S64, int64_t, AV_SAMPLE_FMT_U8,(uint64_t)((*(const uint8_t *) pi - 0x80U))<< 56) CONV_FUNC(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_U8,(*(const uint8_t *) pi - 0x80) *(1.0f/(1<< 7))) CONV_FUNC(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_U8,(*(const uint8_t *) pi - 0x80) *(1.0/(1<< 7))) CONV_FUNC(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S16,(*(const int16_t *) pi >>8)+0x80) CONV_FUNC(AV_SAMPLE_FMT_S64, int64_t, AV_SAMPLE_FMT_S16,(uint64_t)(*(const int16_t *) pi)<< 48) CONV_FUNC(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S16, *(const int16_t *) pi *(1.0f/(1<< 15))) CONV_FUNC(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S16, *(const int16_t *) pi *(1.0/(1<< 15))) CONV_FUNC(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S32,(*(const int32_t *) pi >>24)+0x80) CONV_FUNC(AV_SAMPLE_FMT_S64, int64_t, AV_SAMPLE_FMT_S32,(uint64_t)(*(const int32_t *) pi)<< 32) CONV_FUNC(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S32, *(const int32_t *) pi *(1.0f/(1U<< 31))) CONV_FUNC(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S32, *(const int32_t *) pi *(1.0/(1U<< 31))) CONV_FUNC(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S64,(*(const int64_t *) pi >>56)+0x80) CONV_FUNC(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S64, *(const int64_t *) pi *(1.0f/(INT64_C(1)<< 63))) CONV_FUNC(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S64, *(const int64_t *) pi *(1.0/(INT64_C(1)<< 63))) CONV_FUNC(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_FLT, av_clip_uint8(lrintf(*(const float *) pi *(1<< 7))+0x80)) CONV_FUNC(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_FLT, av_clip_int16(lrintf(*(const float *) pi *(1<< 15)))) CONV_FUNC(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_FLT, av_clipl_int32(llrintf(*(const float *) pi *(1U<< 31)))) CONV_FUNC(AV_SAMPLE_FMT_S64, int64_t, AV_SAMPLE_FMT_FLT, llrintf(*(const float *) pi *(INT64_C(1)<< 63))) CONV_FUNC(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_DBL, av_clip_uint8(lrint(*(const double *) pi *(1<< 7))+0x80)) CONV_FUNC(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_DBL, av_clip_int16(lrint(*(const double *) pi *(1<< 15)))) CONV_FUNC(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_DBL, av_clipl_int32(llrint(*(const double *) pi *(1U<< 31)))) CONV_FUNC(AV_SAMPLE_FMT_S64, int64_t, AV_SAMPLE_FMT_DBL, llrint(*(const double *) pi *(INT64_C(1)<< 63))) #define FMT_PAIR_FUNC(out, in) static conv_func_type *const fmt_pair_to_conv_functions[AV_SAMPLE_FMT_NB *AV_SAMPLE_FMT_NB]={ FMT_PAIR_FUNC(AV_SAMPLE_FMT_U8, AV_SAMPLE_FMT_U8), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_U8), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_U8), FMT_PAIR_FUNC(AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_U8), FMT_PAIR_FUNC(AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_U8), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S64, AV_SAMPLE_FMT_U8), FMT_PAIR_FUNC(AV_SAMPLE_FMT_U8, AV_SAMPLE_FMT_S16), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_S16), FMT_PAIR_FUNC(AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16), FMT_PAIR_FUNC(AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_S16), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S64, AV_SAMPLE_FMT_S16), FMT_PAIR_FUNC(AV_SAMPLE_FMT_U8, AV_SAMPLE_FMT_S32), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S32), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_S32), FMT_PAIR_FUNC(AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S32), FMT_PAIR_FUNC(AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_S32), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S64, AV_SAMPLE_FMT_S32), FMT_PAIR_FUNC(AV_SAMPLE_FMT_U8, AV_SAMPLE_FMT_FLT), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLT), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_FLT), FMT_PAIR_FUNC(AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLT), FMT_PAIR_FUNC(AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_FLT), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S64, AV_SAMPLE_FMT_FLT), FMT_PAIR_FUNC(AV_SAMPLE_FMT_U8, AV_SAMPLE_FMT_DBL), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_DBL), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_DBL), FMT_PAIR_FUNC(AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_DBL), FMT_PAIR_FUNC(AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_DBL), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S64, AV_SAMPLE_FMT_DBL), FMT_PAIR_FUNC(AV_SAMPLE_FMT_U8, AV_SAMPLE_FMT_S64), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S64), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_S64), FMT_PAIR_FUNC(AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S64), FMT_PAIR_FUNC(AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_S64), FMT_PAIR_FUNC(AV_SAMPLE_FMT_S64, AV_SAMPLE_FMT_S64), };static void cpy1(uint8_t **dst, const uint8_t **src, int len){ memcpy(*dst, *src, len);} static void cpy2(uint8_t **dst, const uint8_t **src, int len){ memcpy(*dst, *src, 2 *len);} static void cpy4(uint8_t **dst, const uint8_t **src, int len){ memcpy(*dst, *src, 4 *len);} static void cpy8(uint8_t **dst, const uint8_t **src, int len){ memcpy(*dst, *src, 8 *len);} AudioConvert *swri_audio_convert_alloc(enum AVSampleFormat out_fmt, enum AVSampleFormat in_fmt, int channels, const int *ch_map, int flags) { AudioConvert *ctx;conv_func_type *f=fmt_pair_to_conv_functions[av_get_packed_sample_fmt(out_fmt)+AV_SAMPLE_FMT_NB *av_get_packed_sample_fmt(in_fmt)];if(!f) return NULL;ctx=av_mallocz(sizeof(*ctx));if(!ctx) return NULL;if(channels==1){ in_fmt=av_get_planar_sample_fmt(in_fmt);out_fmt=av_get_planar_sample_fmt(out_fmt);} ctx->channels=channels;ctx->conv_f=f;ctx->ch_map=ch_map;if(in_fmt==AV_SAMPLE_FMT_U8||in_fmt==AV_SAMPLE_FMT_U8P) memset(ctx->silence, 0x80, sizeof(ctx->silence));if(out_fmt==in_fmt &&!ch_map) { switch(av_get_bytes_per_sample(in_fmt)){ case 1:ctx->simd_f=cpy1;break;case 2:ctx->simd_f=cpy2;break;case 4:ctx->simd_f=cpy4;break;case 8:ctx->simd_f=cpy8;break;} } if(HAVE_X86ASM &&HAVE_MMX) swri_audio_convert_init_x86(ctx, out_fmt, in_fmt, channels);if(ARCH_ARM) swri_audio_convert_init_arm(ctx, out_fmt, in_fmt, channels);if(ARCH_AARCH64) swri_audio_convert_init_aarch64(ctx, out_fmt, in_fmt, channels);return ctx;} void swri_audio_convert_free(AudioConvert **ctx) { av_freep(ctx);} int swri_audio_convert(AudioConvert *ctx, AudioData *out, AudioData *in, int len) { int ch;int off=0;const int os=(out->planar ? 1 :out->ch_count) *out->bps;unsigned misaligned=0;av_assert0(ctx->channels==out->ch_count);if(ctx->in_simd_align_mask) { int planes=in->planar ? in->ch_count :1;unsigned m=0;for(ch=0;ch< planes;ch++) m|=(intptr_t) in->ch[ch];misaligned|=m &ctx->in_simd_align_mask;} if(ctx->out_simd_align_mask) { int planes=out->planar ? out->ch_count :1;unsigned m=0;for(ch=0;ch< planes;ch++) m|=(intptr_t) out->ch[ch];misaligned|=m &ctx->out_simd_align_mask;} if(ctx->simd_f &&!ctx->ch_map &&!misaligned){ off=len &~15;av_assert1(off >=0);av_assert1(off<=len);av_assert2(ctx->channels==SWR_CH_MAX||!in->ch[ctx->channels]);if(off >0){ if(out->planar==in->planar){ int planes=out->planar ? out->ch_count :1;for(ch=0;ch< planes;ch++){ ctx->simd_f(out-> ch ch
Definition: audioconvert.c:56
static float calc_reduction_3gpp(float a, float desired_pe, float pe, float active_lines)
Definition: aacpsy.c:560
float ath
absolute threshold of hearing per bands
Definition: aacpsy.c:141
#define PSY_3GPP_EN_SPREAD_HI_L1
Definition: aacpsy.c:47
static av_cold float ath(float f, float add)
Calculate ATH value for given frequency.
Definition: aacpsy.c:292
float prev_energy_subshort[AAC_NUM_BLOCKS_SHORT *PSY_LAME_NUM_SUBBLOCKS]
Definition: aacpsy.c:133
enum WindowSequence next_window_seq
window sequence to be used in the next frame
Definition: aacpsy.c:130
#define PSY_SNR_25DB
Definition: aacpsy.c:65
#define AAC_BLOCK_SIZE_LONG
long block size
Definition: aacpsy.c:96
int * num_bands
number of scalefactor bands for possible frame sizes
Definition: psymodel.h:99
Macro definitions for various function/variable attributes.
LAME psy model preset struct.
Definition: aacpsy.c:169
float thr
energy threshold
Definition: aacpsy.c:110
float correction
PE correction factor.
Definition: aacpsy.c:159
static av_cold void psy_3gpp_end(FFPsyContext *apc)
Definition: aacpsy.c:858
float attack_threshold
attack threshold for this channel
Definition: aacpsy.c:132
#define PSY_3GPP_EN_SPREAD_LOW_L
Definition: aacpsy.c:53
float nz_lines
number of non-zero spectral lines
Definition: aacpsy.c:112
uint8_t
psychoacoustic model frame type-dependent coefficients
Definition: aacpsy.c:140
#define av_cold
Definition: attributes.h:82
int size
size of the bitresevoir in bits
Definition: psymodel.h:103
#define f(width, name)
Definition: cbs_vp9.c:255
static float calc_reduced_thr_3gpp(AacPsyBand *band, float min_snr, float reduction)
Definition: aacpsy.c:574
#define PSY_3GPP_C2
Definition: aacpsy.c:61
#define PSY_LAME_FIR_LEN
LAME psy model FIR order.
Definition: aacpsy.c:95
#define PSY_3GPP_CLIP_LO_L
Definition: aacpsy.c:75
#define PSY_3GPP_SPEND_SLOPE_S
Definition: aacpsy.c:72
#define u(width, name, range_min, range_max)
Definition: cbs_h2645.c:253
#define PSY_3GPP_THR_SPREAD_LOW
Definition: aacpsy.c:45
context used by psychoacoustic model
Definition: psymodel.h:89
#define atanf(x)
Definition: libm.h:40
int flags
Flags modifying the (de)muxer behaviour.
Definition: avformat.h:1482
struct FFPsyContext::@127 bitres
#define AAC_CUTOFF(s)
Definition: psymodel.h:41
single band psychoacoustic information
Definition: psymodel.h:50
static float lame_calc_attack_threshold(int bitrate)
Calculate the ABR attack threshold from the above LAME psymodel table.
Definition: aacpsy.c:232
uint8_t next_grouping
stored grouping scheme for the next frame (in case of 8 short window sequence)
Definition: aacpsy.c:129
#define PSY_3GPP_SAVE_ADD_L
Definition: aacpsy.c:69
static av_cold float calc_bark(float f)
Calculate Bark value for given line.
Definition: aacpsy.c:282
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:258
static av_always_inline double ff_exp10(double x)
Compute 10^x for floating point values.
Definition: ffmath.h:42
#define AVERROR(e)
Definition: error.h:43
#define PSY_3GPP_SPEND_ADD_S
Definition: aacpsy.c:74
#define PSY_SNR_1DB
Definition: aacpsy.c:64
3GPP TS26.403-inspired psychoacoustic model specific data
Definition: aacpsy.c:151
single/pair channel context for psychoacoustic model
Definition: aacpsy.c:123
static const float psy_fir_coeffs[]
LAME psy model FIR coefficient table.
Definition: aacpsy.c:219
int bits
Definition: psymodel.h:51
float barks
Bark value for each spectral band in long frame.
Definition: aacpsy.c:142
int flags
AV_CODEC_FLAG_*.
Definition: avcodec.h:1613
float pe_const
constant part of the PE calculation
Definition: aacpsy.c:115
void * av_mallocz(size_t size)
Allocate a memory block with alignment suitable for all memory accesses (including vectors if availab...
Definition: mem.c:236
int num_windows
number of windows in a frame
Definition: psymodel.h:80
static FFPsyWindowInfo psy_lame_window(FFPsyContext *ctx, const float *audio, const float *la, int channel, int prev_type)
Definition: aacpsy.c:883
#define PSY_3GPP_SPEND_SLOPE_L
Definition: aacpsy.c:71
#define PSY_3GPP_THR_SPREAD_HI
constants for 3GPP AAC psychoacoustic model
Definition: aacpsy.c:44
float energy
Definition: psymodel.h:52
WindowSequence
Definition: aac.h:75
#define FFMAX(a, b)
Definition: common.h:94
codec-specific psychoacoustic model implementation
Definition: psymodel.h:114
#define PSY_3GPP_RPELEV
Definition: aacpsy.c:58
int8_t exp
Definition: eval.c:72
struct AacPsyContext::@31 pe
float thr_quiet
threshold in quiet
Definition: aacpsy.c:111
static struct @303 state
static void psy_3gpp_analyze(FFPsyContext *ctx, int channel, const float **coeffs, const FFPsyWindowInfo *wi)
Definition: aacpsy.c:848
#define AV_CODEC_FLAG_QSCALE
Use fixed qscale.
Definition: avcodec.h:842
#define NAN
Definition: mathematics.h:64
#define FFMIN(a, b)
Definition: common.h:96
int prev_attack
attack value for the last short block in the previous sequence
Definition: aacpsy.c:134
#define PSY_3GPP_SAVE_SLOPE_S
Definition: aacpsy.c:68
#define PSY_3GPP_C3
Definition: aacpsy.c:62
uint8_t w
Definition: llviddspenc.c:38
uint8_t num_ch
number of channels in this group
Definition: psymodel.h:70
int frame_bits
average bits per frame
Definition: aacpsy.c:153
int fill_level
bit reservoir fill level
Definition: aacpsy.c:154
AVFormatContext * ctx
Definition: movenc.c:48
static void lame_apply_block_type(AacPsyChannel *ctx, FFPsyWindowInfo *wi, int uselongblock)
Definition: aacpsy.c:865
#define PSY_3GPP_SAVE_SLOPE_L
Definition: aacpsy.c:67
#define s(width, name)
Definition: cbs_vp9.c:257
Reference: libavcodec/aacpsy.c.
#define PSY_LAME_NUM_SUBBLOCKS
Number of sub-blocks in each short block.
Definition: aacpsy.c:99
#define ATH_ADD
Definition: aacpsy.c:287
float energy
band energy
Definition: aacpsy.c:109
const FFPsyModel ff_aac_psy_model
Definition: aacpsy.c:1021
static void psy_3gpp_analyze_channel(FFPsyContext *ctx, int channel, const float *coefs, const FFPsyWindowInfo *wi)
Calculate band thresholds as suggested in 3GPP TS26.403.
Definition: aacpsy.c:652
float st_lrm
short threshold for L, R, and M channels
Definition: aacpsy.c:174
#define PSY_3GPP_EN_SPREAD_LOW_S
Definition: aacpsy.c:55
#define exp2f(x)
Definition: libm.h:293
Libavcodec external API header.
int sample_rate
samples per second
Definition: avcodec.h:2189
FFPsyChannelGroup * ff_psy_find_group(FFPsyContext *ctx, int channel)
Determine what group a channel belongs to.
Definition: psymodel.c:73
main external API structure.
Definition: avcodec.h:1533
float win_energy
sliding average of channel energy
Definition: aacpsy.c:127
void * model_priv_data
psychoacoustic model implementation private data
Definition: psymodel.h:108
float active_lines
number of active spectral lines
Definition: aacpsy.c:113
static const float bands[]
static float iir_filter(int in, float state[2])
IIR filter used in block switching decision.
Definition: aacpsy.c:387
int avoid_holes
hole avoidance flag
Definition: aacpsy.c:117
AacPsyBand band[128]
bands information
Definition: aacpsy.c:124
uint8_t pi<< 24) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_U8, uint8_t,(*(const uint8_t *) pi - 0x80) *(1.0f/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_U8, uint8_t,(*(const uint8_t *) pi - 0x80) *(1.0/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S16, int16_t,(*(const int16_t *) pi >> 8)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S16, int16_t, *(const int16_t *) pi *(1.0f/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S16, int16_t, *(const int16_t *) pi *(1.0/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S32, int32_t,(*(const int32_t *) pi >> 24)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S32, int32_t, *(const int32_t *) pi *(1.0f/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S32, int32_t, *(const int32_t *) pi *(1.0/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_FLT, float, av_clip_uint8(lrintf(*(const float *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_FLT, float, av_clip_int16(lrintf(*(const float *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_FLT, float, av_clipl_int32(llrintf(*(const float *) pi *(1U<< 31)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_DBL, double, av_clip_uint8(lrint(*(const double *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_DBL, double, av_clip_int16(lrint(*(const double *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_DBL, double, av_clipl_int32(llrint(*(const double *) pi *(1U<< 31)))) #define SET_CONV_FUNC_GROUP(ofmt, ifmt) static void set_generic_function(AudioConvert *ac) { } void ff_audio_convert_free(AudioConvert **ac) { if(! *ac) return;ff_dither_free(&(*ac) ->dc);av_freep(ac);} AudioConvert *ff_audio_convert_alloc(AVAudioResampleContext *avr, enum AVSampleFormat out_fmt, enum AVSampleFormat in_fmt, int channels, int sample_rate, int apply_map) { AudioConvert *ac;int in_planar, out_planar;ac=av_mallocz(sizeof(*ac));if(!ac) return NULL;ac->avr=avr;ac->out_fmt=out_fmt;ac->in_fmt=in_fmt;ac->channels=channels;ac->apply_map=apply_map;if(avr->dither_method !=AV_RESAMPLE_DITHER_NONE &&av_get_packed_sample_fmt(out_fmt)==AV_SAMPLE_FMT_S16 &&av_get_bytes_per_sample(in_fmt) > 2) { ac->dc=ff_dither_alloc(avr, out_fmt, in_fmt, channels, sample_rate, apply_map);if(!ac->dc) { av_free(ac);return NULL;} return ac;} in_planar=ff_sample_fmt_is_planar(in_fmt, channels);out_planar=ff_sample_fmt_is_planar(out_fmt, channels);if(in_planar==out_planar) { ac->func_type=CONV_FUNC_TYPE_FLAT;ac->planes=in_planar ? ac->channels :1;} else if(in_planar) ac->func_type=CONV_FUNC_TYPE_INTERLEAVE;else ac->func_type=CONV_FUNC_TYPE_DEINTERLEAVE;set_generic_function(ac);if(ARCH_AARCH64) ff_audio_convert_init_aarch64(ac);if(ARCH_ARM) ff_audio_convert_init_arm(ac);if(ARCH_X86) ff_audio_convert_init_x86(ac);return ac;} int ff_audio_convert(AudioConvert *ac, AudioData *out, AudioData *in) { int use_generic=1;int len=in->nb_samples;int p;if(ac->dc) { av_log(ac->avr, AV_LOG_TRACE, "%d samples - audio_convert: %s to %s (dithered)\", len, av_get_sample_fmt_name(ac->in_fmt), av_get_sample_fmt_name(ac->out_fmt));return ff_convert_dither(ac-> in
#define PSY_3GPP_CLIP_HI_S
Definition: aacpsy.c:78
#define PSY_3GPP_RPEMIN
Definition: aacpsy.c:57
static const PsyLamePreset psy_abr_map[]
LAME psy model preset table for ABR.
Definition: aacpsy.c:180
int window_shape
window shape (sine/KBD/whatever)
Definition: psymodel.h:79
#define PSY_PE_FORGET_SLOPE
Definition: aacpsy.c:83
#define PSY_3GPP_PE_TO_BITS(bits)
Definition: aacpsy.c:92
int cutoff
lowpass frequency cutoff for analysis
Definition: psymodel.h:96
float min_snr
minimal SNR
Definition: aacpsy.c:145
float max
maximum allowed PE for bit factor calculation
Definition: aacpsy.c:157
static void calc_thr_3gpp(const FFPsyWindowInfo *wi, const int num_bands, AacPsyChannel *pch, const uint8_t *band_sizes, const float *coefs, const int cutoff)
Definition: aacpsy.c:600
float previous
allowed PE of the previous frame
Definition: aacpsy.c:158
AacPsyCoeffs psy_coef[2][64]
Definition: aacpsy.c:161
float min
minimum allowed PE for bit factor calculation
Definition: aacpsy.c:156
int global_quality
Global quality for codecs which cannot change it per frame.
Definition: avcodec.h:1599
static av_cold int psy_3gpp_init(FFPsyContext *ctx)
Definition: aacpsy.c:301
static void psy_hp_filter(const float *firbuf, float *hpfsmpl, const float *psy_fir_coeffs)
Definition: aacpsy.c:631
float spread_hi[2]
spreading factor for high-to-low threshold spreading in long frame
Definition: aacpsy.c:144
const char * name
Definition: psymodel.h:115
int64_t bitrate
Definition: h264_levels.c:89
internal math functions header
static av_unused FFPsyWindowInfo psy_3gpp_window(FFPsyContext *ctx, const int16_t *audio, const int16_t *la, int channel, int prev_type)
Tell encoder which window types to use.
Definition: aacpsy.c:408
int
static float calc_pe_3gpp(AacPsyBand *band)
Definition: aacpsy.c:537
#define exp2(x)
Definition: libm.h:288
windowing related information
Definition: psymodel.h:77
#define log2f(x)
Definition: libm.h:409
channel
Use these values when setting the channel map with ebur128_set_channel().
Definition: ebur128.h:39
#define PSY_3GPP_BITS_TO_PE(bits)
Definition: aacpsy.c:91
#define PSY_3GPP_C1
Definition: aacpsy.c:60
float norm_fac
normalization factor for linearization
Definition: aacpsy.c:116
int chan_bitrate
bitrate per channel
Definition: aacpsy.c:152
#define PSY_3GPP_CLIP_LO_S
Definition: aacpsy.c:76
#define PSY_3GPP_AH_THR_LONG
Definition: aacpsy.c:80
int channels
number of audio channels
Definition: avcodec.h:2190
float pe
perceptual entropy
Definition: aacpsy.c:114
#define PSY_3GPP_EN_SPREAD_HI_S
Definition: aacpsy.c:51
static const double coeff[2][5]
Definition: vf_owdenoise.c:72
#define FF_QP2LAMBDA
factor to convert from H.263 QP to lambda
Definition: avutil.h:227
AacPsyChannel * ch
Definition: aacpsy.c:162
#define PSY_3GPP_SAVE_ADD_S
Definition: aacpsy.c:70
#define av_freep(p)
void INT64 start
Definition: avisynth_c.h:690
information for single band used by 3GPP TS26.403-inspired psychoacoustic model
Definition: aacpsy.c:108
AVCodecContext * avctx
encoder context
Definition: psymodel.h:90
float threshold
Definition: psymodel.h:53
AAC data declarations.
float spread_low[2]
spreading factor for low-to-high threshold spreading in long frame
Definition: aacpsy.c:143
#define PSY_3GPP_CLIP_HI_L
Definition: aacpsy.c:77
float spread
Definition: psymodel.h:54
int window_type[3]
window type (short/long/transitional, etc.) - current, previous and next
Definition: psymodel.h:78
#define AAC_NUM_BLOCKS_SHORT
number of blocks in a short sequence
Definition: aacpsy.c:98
#define av_unused
Definition: attributes.h:125
#define PSY_3GPP_SPEND_ADD_L
Definition: aacpsy.c:73
void * av_mallocz_array(size_t nmemb, size_t size)
Allocate a memory block for an array with av_mallocz().
Definition: mem.c:191
static av_cold void lame_window_init(AacPsyContext *ctx, AVCodecContext *avctx)
LAME psy model specific initialization.
Definition: aacpsy.c:262