FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
aacenc_is.c
Go to the documentation of this file.
1 /*
2  * AAC encoder intensity stereo
3  * Copyright (C) 2015 Rostislav Pehlivanov
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 Intensity Stereo
25  * @author Rostislav Pehlivanov ( atomnuker gmail com )
26  */
27 
28 #include "aacenc.h"
29 #include "aacenc_utils.h"
30 #include "aacenc_is.h"
31 #include "aacenc_quantization.h"
32 
34  int start, int w, int g, float ener0,
35  float ener1, float ener01,
36  int use_pcoeffs, int phase)
37 {
38  int i, w2;
39  SingleChannelElement *sce0 = &cpe->ch[0];
40  SingleChannelElement *sce1 = &cpe->ch[1];
41  float *L = use_pcoeffs ? sce0->pcoeffs : sce0->coeffs;
42  float *R = use_pcoeffs ? sce1->pcoeffs : sce1->coeffs;
43  float *L34 = &s->scoefs[256*0], *R34 = &s->scoefs[256*1];
44  float *IS = &s->scoefs[256*2], *I34 = &s->scoefs[256*3];
45  float dist1 = 0.0f, dist2 = 0.0f;
46  struct AACISError is_error = {0};
47 
48  if (ener01 <= 0 || ener0 <= 0) {
49  is_error.pass = 0;
50  return is_error;
51  }
52 
53  for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
54  FFPsyBand *band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)*16+g];
55  FFPsyBand *band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)*16+g];
56  int is_band_type, is_sf_idx = FFMAX(1, sce0->sf_idx[w*16+g]-4);
57  float e01_34 = phase*pos_pow34(ener1/ener0);
58  float maxval, dist_spec_err = 0.0f;
59  float minthr = FFMIN(band0->threshold, band1->threshold);
60  for (i = 0; i < sce0->ics.swb_sizes[g]; i++)
61  IS[i] = (L[start+(w+w2)*128+i] + phase*R[start+(w+w2)*128+i])*sqrt(ener0/ener01);
62  s->abs_pow34(L34, &L[start+(w+w2)*128], sce0->ics.swb_sizes[g]);
63  s->abs_pow34(R34, &R[start+(w+w2)*128], sce0->ics.swb_sizes[g]);
64  s->abs_pow34(I34, IS, sce0->ics.swb_sizes[g]);
65  maxval = find_max_val(1, sce0->ics.swb_sizes[g], I34);
66  is_band_type = find_min_book(maxval, is_sf_idx);
67  dist1 += quantize_band_cost(s, &L[start + (w+w2)*128], L34,
68  sce0->ics.swb_sizes[g],
69  sce0->sf_idx[w*16+g],
70  sce0->band_type[w*16+g],
71  s->lambda / band0->threshold, INFINITY, NULL, NULL, 0);
72  dist1 += quantize_band_cost(s, &R[start + (w+w2)*128], R34,
73  sce1->ics.swb_sizes[g],
74  sce1->sf_idx[w*16+g],
75  sce1->band_type[w*16+g],
76  s->lambda / band1->threshold, INFINITY, NULL, NULL, 0);
77  dist2 += quantize_band_cost(s, IS, I34, sce0->ics.swb_sizes[g],
78  is_sf_idx, is_band_type,
79  s->lambda / minthr, INFINITY, NULL, NULL, 0);
80  for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
81  dist_spec_err += (L34[i] - I34[i])*(L34[i] - I34[i]);
82  dist_spec_err += (R34[i] - I34[i]*e01_34)*(R34[i] - I34[i]*e01_34);
83  }
84  dist_spec_err *= s->lambda / minthr;
85  dist2 += dist_spec_err;
86  }
87 
88  is_error.pass = dist2 <= dist1;
89  is_error.phase = phase;
90  is_error.error = dist2 - dist1;
91  is_error.dist1 = dist1;
92  is_error.dist2 = dist2;
93  is_error.ener01 = ener01;
94 
95  return is_error;
96 }
97 
99 {
100  SingleChannelElement *sce0 = &cpe->ch[0];
101  SingleChannelElement *sce1 = &cpe->ch[1];
102  int start = 0, count = 0, w, w2, g, i, prev_sf1 = -1, prev_bt = -1, prev_is = 0;
103  const float freq_mult = avctx->sample_rate/(1024.0f/sce0->ics.num_windows)/2.0f;
104  uint8_t nextband1[128];
105 
106  if (!cpe->common_window)
107  return;
108 
109  /** Scout out next nonzero bands */
110  ff_init_nextband_map(sce1, nextband1);
111 
112  for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) {
113  start = 0;
114  for (g = 0; g < sce0->ics.num_swb; g++) {
115  if (start*freq_mult > INT_STEREO_LOW_LIMIT*(s->lambda/170.0f) &&
116  cpe->ch[0].band_type[w*16+g] != NOISE_BT && !cpe->ch[0].zeroes[w*16+g] &&
117  cpe->ch[1].band_type[w*16+g] != NOISE_BT && !cpe->ch[1].zeroes[w*16+g] &&
118  ff_sfdelta_can_remove_band(sce1, nextband1, prev_sf1, w*16+g)) {
119  float ener0 = 0.0f, ener1 = 0.0f, ener01 = 0.0f, ener01p = 0.0f;
120  struct AACISError ph_err1, ph_err2, *best;
121  for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
122  for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
123  float coef0 = sce0->coeffs[start+(w+w2)*128+i];
124  float coef1 = sce1->coeffs[start+(w+w2)*128+i];
125  ener0 += coef0*coef0;
126  ener1 += coef1*coef1;
127  ener01 += (coef0 + coef1)*(coef0 + coef1);
128  ener01p += (coef0 - coef1)*(coef0 - coef1);
129  }
130  }
131  ph_err1 = ff_aac_is_encoding_err(s, cpe, start, w, g,
132  ener0, ener1, ener01p, 0, -1);
133  ph_err2 = ff_aac_is_encoding_err(s, cpe, start, w, g,
134  ener0, ener1, ener01, 0, +1);
135  best = (ph_err1.pass && ph_err1.error < ph_err2.error) ? &ph_err1 : &ph_err2;
136  if (best->pass) {
137  cpe->is_mask[w*16+g] = 1;
138  cpe->ms_mask[w*16+g] = 0;
139  cpe->ch[0].is_ener[w*16+g] = sqrt(ener0 / best->ener01);
140  cpe->ch[1].is_ener[w*16+g] = ener0/ener1;
141  cpe->ch[1].band_type[w*16+g] = (best->phase > 0) ? INTENSITY_BT : INTENSITY_BT2;
142  if (prev_is && prev_bt != cpe->ch[1].band_type[w*16+g]) {
143  /** Flip M/S mask and pick the other CB, since it encodes more efficiently */
144  cpe->ms_mask[w*16+g] = 1;
145  cpe->ch[1].band_type[w*16+g] = (best->phase > 0) ? INTENSITY_BT2 : INTENSITY_BT;
146  }
147  prev_bt = cpe->ch[1].band_type[w*16+g];
148  count++;
149  }
150  }
151  if (!sce1->zeroes[w*16+g] && sce1->band_type[w*16+g] < RESERVED_BT)
152  prev_sf1 = sce1->sf_idx[w*16+g];
153  prev_is = cpe->is_mask[w*16+g];
154  start += sce0->ics.swb_sizes[g];
155  }
156  }
157  cpe->is_mode = !!count;
158 }
#define NULL
Definition: coverity.c:32
const char * s
Definition: avisynth_c.h:768
Band types following are encoded differently from others.
Definition: aac.h:86
Definition: vf_geq.c:46
float dist2
Definition: aacenc_is.h:41
const char * g
Definition: vf_curves.c:112
static int ff_sfdelta_can_remove_band(const SingleChannelElement *sce, const uint8_t *nextband, int prev_sf, int band)
Definition: aacenc_utils.h:232
int common_window
Set if channels share a common 'IndividualChannelStream' in bitstream.
Definition: aac.h:278
uint8_t ms_mask[128]
Set if mid/side stereo is used for each scalefactor window band.
Definition: aac.h:281
float lambda
Definition: aacenc.h:117
Spectral data are scaled white noise not coded in the bitstream.
Definition: aac.h:87
AAC encoder quantizer.
INTFLOAT pcoeffs[1024]
coefficients for IMDCT, pristine
Definition: aac.h:261
AAC encoder context.
Definition: aacenc.h:95
uint8_t
SingleChannelElement ch[2]
Definition: aac.h:284
float ener01
Definition: aacenc_is.h:42
int pass
Definition: aacenc_is.h:37
Scalefactor data are intensity stereo positions (in phase).
Definition: aac.h:89
single band psychoacoustic information
Definition: psymodel.h:50
float dist1
Definition: aacenc_is.h:40
float is_ener[128]
Intensity stereo pos (used by encoder)
Definition: aac.h:259
GLsizei count
Definition: opengl_enc.c:109
int num_swb
number of scalefactor window bands
Definition: aac.h:183
#define FFMAX(a, b)
Definition: common.h:94
float error
Definition: aacenc_is.h:39
#define FFMIN(a, b)
Definition: common.h:96
static void ff_init_nextband_map(const SingleChannelElement *sce, uint8_t *nextband)
Definition: aacenc_utils.h:199
AAC encoder Intensity Stereo.
#define L(x)
Definition: vp56_arith.h:36
int phase
Definition: aacenc_is.h:38
#define INFINITY
Definition: math.h:27
void ff_aac_search_for_is(AACEncContext *s, AVCodecContext *avctx, ChannelElement *cpe)
Definition: aacenc_is.c:98
static int find_min_book(float maxval, int sf)
Definition: aacenc_utils.h:92
int sample_rate
samples per second
Definition: avcodec.h:2438
main external API structure.
Definition: avcodec.h:1676
IndividualChannelStream ics
Definition: aac.h:249
uint8_t group_len[8]
Definition: aac.h:179
static float pos_pow34(float a)
Definition: aacenc_utils.h:49
const uint8_t * swb_sizes
table of scalefactor band sizes for a particular window
Definition: aac.h:182
uint8_t zeroes[128]
band is not coded (used by encoder)
Definition: aac.h:257
int sf_idx[128]
scalefactor indices (used by encoder)
Definition: aac.h:256
uint8_t is_mode
Set if any bands have been encoded using intensity stereo (used by encoder)
Definition: aac.h:280
INTFLOAT coeffs[1024]
coefficients for IMDCT, maybe processed
Definition: aac.h:262
Scalefactor data are intensity stereo positions (out of phase).
Definition: aac.h:88
AAC encoder utilities.
Single Channel Element - used for both SCE and LFE elements.
Definition: aac.h:248
struct AACISError ff_aac_is_encoding_err(AACEncContext *s, ChannelElement *cpe, int start, int w, int g, float ener0, float ener1, float ener01, int use_pcoeffs, int phase)
Definition: aacenc_is.c:33
channel element - generic struct for SCE/CPE/CCE/LFE
Definition: aac.h:275
enum BandType band_type[128]
band types
Definition: aac.h:252
#define INT_STEREO_LOW_LIMIT
Frequency in Hz for lower limit of intensity stereo.
Definition: aacenc_is.h:34
static float find_max_val(int group_len, int swb_size, const float *scaled)
Definition: aacenc_utils.h:80
void INT64 start
Definition: avisynth_c.h:690
static float quantize_band_cost(struct AACEncContext *s, const float *in, const float *scaled, int size, int scale_idx, int cb, const float lambda, const float uplim, int *bits, float *energy, int rtz)
uint8_t is_mask[128]
Set if intensity stereo is used (used by encoder)
Definition: aac.h:282
float threshold
Definition: psymodel.h:53