FFmpeg
aacenc_ltp.c
Go to the documentation of this file.
1 /*
2  * AAC encoder long term prediction extension
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 long term prediction extension
25  * @author Rostislav Pehlivanov ( atomnuker gmail com )
26  */
27 
28 #include "aacenc_ltp.h"
29 #include "aacenc_quantization.h"
30 #include "aacenc_utils.h"
31 
32 /**
33  * Encode LTP data.
34  */
36  int common_window)
37 {
38  int i;
39  IndividualChannelStream *ics = &sce->ics;
40  if (s->profile != AV_PROFILE_AAC_LTP || !ics->predictor_present)
41  return;
42  if (common_window)
43  put_bits(&s->pb, 1, 0);
44  put_bits(&s->pb, 1, ics->ltp.present);
45  if (!ics->ltp.present)
46  return;
47  put_bits(&s->pb, 11, ics->ltp.lag);
48  put_bits(&s->pb, 3, ics->ltp.coef_idx);
49  for (i = 0; i < FFMIN(ics->max_sfb, MAX_LTP_LONG_SFB); i++)
50  put_bits(&s->pb, 1, ics->ltp.used[i]);
51 }
52 
54 {
55  int i, ch, tag, chans, cur_channel, start_ch = 0;
56  ChannelElement *cpe;
58  for (i = 0; i < s->chan_map[0]; i++) {
59  cpe = &s->cpe[i];
60  tag = s->chan_map[i+1];
61  chans = tag == TYPE_CPE ? 2 : 1;
62  for (ch = 0; ch < chans; ch++) {
63  sce = &cpe->ch[ch];
64  cur_channel = start_ch + ch;
65  /* New sample + overlap */
66  memcpy(&sce->ltp_state[0], &sce->ltp_state[1024], 1024*sizeof(sce->ltp_state[0]));
67  memcpy(&sce->ltp_state[1024], &s->planar_samples[cur_channel][2048], 1024*sizeof(sce->ltp_state[0]));
68  memcpy(&sce->ltp_state[2048], &sce->ret_buf[0], 1024*sizeof(sce->ltp_state[0]));
69  sce->ics.ltp.lag = 0;
70  }
71  start_ch += chans;
72  }
73 }
74 
75 static void get_lag(float *buf, const float *new, LongTermPrediction *ltp)
76 {
77  int i, j, lag = 0, max_corr = 0;
78  float max_ratio = 0.0f;
79  for (i = 0; i < 2048; i++) {
80  float corr, s0 = 0.0f, s1 = 0.0f;
81  const int start = FFMAX(0, i - 1024);
82  for (j = start; j < 2048; j++) {
83  const int idx = j - i + 1024;
84  s0 += new[j]*buf[idx];
85  s1 += buf[idx]*buf[idx];
86  }
87  corr = s1 > 0.0f ? s0/sqrt(s1) : 0.0f;
88  if (corr > max_corr) {
89  max_corr = corr;
90  lag = i;
91  max_ratio = corr/(2048-start);
92  }
93  }
94  ltp->lag = FFMAX(av_clip_uintp2(lag, 11), 0);
95  ltp->coef_idx = quant_array_idx(max_ratio, ff_ltp_coef, 8);
96  ltp->coef = ff_ltp_coef[ltp->coef_idx];
97 }
98 
99 static void generate_samples(float *buf, LongTermPrediction *ltp)
100 {
101  int i, samples_num = 2048;
102  if (!ltp->lag) {
103  ltp->present = 0;
104  return;
105  } else if (ltp->lag < 1024) {
106  samples_num = ltp->lag + 1024;
107  }
108  for (i = 0; i < samples_num; i++)
109  buf[i] = ltp->coef*buf[i + 2048 - ltp->lag];
110  memset(&buf[i], 0, (2048 - i)*sizeof(float));
111 }
112 
113 /**
114  * Process LTP parameters
115  * @see Patent WO2006070265A1
116  */
118 {
119  float *pred_signal = &sce->ltp_state[0];
120  const float *samples = &s->planar_samples[s->cur_channel][1024];
121 
122  if (s->profile != AV_PROFILE_AAC_LTP)
123  return;
124 
125  /* Calculate lag */
126  get_lag(pred_signal, samples, &sce->ics.ltp);
127  generate_samples(pred_signal, &sce->ics.ltp);
128 }
129 
131 {
132  int sfb, count = 0;
133  SingleChannelElement *sce0 = &cpe->ch[0];
134  SingleChannelElement *sce1 = &cpe->ch[1];
135 
136  if (!cpe->common_window ||
139  sce0->ics.ltp.present = 0;
140  return;
141  }
142 
143  for (sfb = 0; sfb < FFMIN(sce0->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++) {
144  int sum = sce0->ics.ltp.used[sfb] + sce1->ics.ltp.used[sfb];
145  if (sum != 2) {
146  sce0->ics.ltp.used[sfb] = 0;
147  } else {
148  count++;
149  }
150  }
151 
152  sce0->ics.ltp.present = !!count;
153  sce0->ics.predictor_present = !!count;
154 }
155 
156 /**
157  * Mark LTP sfb's
158  */
160  int common_window)
161 {
162  int w, g, w2, i, start = 0, count = 0;
163  int saved_bits = -(15 + FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB));
164  float *C34 = &s->scoefs[128*0], *PCD = &s->scoefs[128*1];
165  float *PCD34 = &s->scoefs[128*2];
166  const int max_ltp = FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB);
167 
168  if (sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
169  if (sce->ics.ltp.lag) {
170  memset(&sce->ltp_state[0], 0, 3072*sizeof(sce->ltp_state[0]));
171  memset(&sce->ics.ltp, 0, sizeof(LongTermPrediction));
172  }
173  return;
174  }
175 
176  if (!sce->ics.ltp.lag || s->lambda > 120.0f)
177  return;
178 
179  for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
180  start = 0;
181  for (g = 0; g < sce->ics.num_swb; g++) {
182  int bits1 = 0, bits2 = 0;
183  float dist1 = 0.0f, dist2 = 0.0f;
184  if (w*16+g > max_ltp) {
185  start += sce->ics.swb_sizes[g];
186  continue;
187  }
188  for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
189  int bits_tmp1, bits_tmp2;
190  FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
191  for (i = 0; i < sce->ics.swb_sizes[g]; i++)
192  PCD[i] = sce->coeffs[start+(w+w2)*128+i] - sce->lcoeffs[start+(w+w2)*128+i];
193  s->aacdsp.abs_pow34(C34, &sce->coeffs[start+(w+w2)*128], sce->ics.swb_sizes[g]);
194  s->aacdsp.abs_pow34(PCD34, PCD, sce->ics.swb_sizes[g]);
195  dist1 += quantize_band_cost(s, &sce->coeffs[start+(w+w2)*128], C34, sce->ics.swb_sizes[g],
196  sce->sf_idx[(w+w2)*16+g], sce->band_type[(w+w2)*16+g],
197  s->lambda/band->threshold, INFINITY, &bits_tmp1, NULL);
198  dist2 += quantize_band_cost(s, PCD, PCD34, sce->ics.swb_sizes[g],
199  sce->sf_idx[(w+w2)*16+g],
200  sce->band_type[(w+w2)*16+g],
201  s->lambda/band->threshold, INFINITY, &bits_tmp2, NULL);
202  bits1 += bits_tmp1;
203  bits2 += bits_tmp2;
204  }
205  if (dist2 < dist1 && bits2 < bits1) {
206  for (w2 = 0; w2 < sce->ics.group_len[w]; w2++)
207  for (i = 0; i < sce->ics.swb_sizes[g]; i++)
208  sce->coeffs[start+(w+w2)*128+i] -= sce->lcoeffs[start+(w+w2)*128+i];
209  sce->ics.ltp.used[w*16+g] = 1;
210  saved_bits += bits1 - bits2;
211  count++;
212  }
213  start += sce->ics.swb_sizes[g];
214  }
215  }
216 
217  sce->ics.ltp.present = !!count && (saved_bits >= 0);
218  sce->ics.predictor_present = !!sce->ics.ltp.present;
219 
220  /* Reset any marked sfbs */
221  if (!sce->ics.ltp.present && !!count) {
222  for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
223  start = 0;
224  for (g = 0; g < sce->ics.num_swb; g++) {
225  if (sce->ics.ltp.used[w*16+g]) {
226  for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
227  for (i = 0; i < sce->ics.swb_sizes[g]; i++) {
228  sce->coeffs[start+(w+w2)*128+i] += sce->lcoeffs[start+(w+w2)*128+i];
229  }
230  }
231  }
232  start += sce->ics.swb_sizes[g];
233  }
234  }
235  }
236 }
SingleChannelElement::lcoeffs
float lcoeffs[1024]
MDCT of LTP coefficients.
Definition: aacenc.h:128
INFINITY
#define INFINITY
Definition: mathematics.h:118
AACISError::dist2
float dist2
Definition: aacenc_is.h:41
av_clip_uintp2
#define av_clip_uintp2
Definition: common.h:122
ff_aac_update_ltp
void ff_aac_update_ltp(AACEncContext *s, SingleChannelElement *sce)
Process LTP parameters.
Definition: aacenc_ltp.c:117
get_lag
static void get_lag(float *buf, const float *new, LongTermPrediction *ltp)
Definition: aacenc_ltp.c:75
LongTermPrediction::used
int8_t used[MAX_LTP_LONG_SFB]
Definition: aacdec.h:78
put_bits
static void put_bits(Jpeg2000EncoderContext *s, int val, int n)
put n times val bit
Definition: j2kenc.c:222
aacenc_ltp.h
w
uint8_t w
Definition: llviddspenc.c:38
ff_ltp_coef
const float ff_ltp_coef[8]
FFMAX
#define FFMAX(a, b)
Definition: macros.h:47
ChannelElement::ch
SingleChannelElement ch[2]
Definition: aacdec.h:153
IndividualChannelStream::num_swb
int num_swb
number of scalefactor window bands
Definition: aacdec.h:92
AACISError::dist1
float dist1
Definition: aacenc_is.h:40
LongTermPrediction::coef
INTFLOAT coef
Definition: aacdec.h:77
SingleChannelElement::ret_buf
INTFLOAT ret_buf[2048]
PCM output buffer.
Definition: aacdec.h:139
TYPE_CPE
@ TYPE_CPE
Definition: aac.h:44
MAX_LTP_LONG_SFB
#define MAX_LTP_LONG_SFB
Definition: aac.h:40
SingleChannelElement::ics
IndividualChannelStream ics
Definition: aacdec.h:132
s
#define s(width, name)
Definition: cbs_vp9.c:198
SingleChannelElement::coeffs
INTFLOAT coeffs[1024]
coefficients for IMDCT, maybe processed
Definition: aacdec.h:137
bits1
static const uint8_t bits1[81]
Definition: aactab.c:221
AV_PROFILE_AAC_LTP
#define AV_PROFILE_AAC_LTP
Definition: defs.h:71
IndividualChannelStream::swb_sizes
const uint8_t * swb_sizes
table of scalefactor band sizes for a particular window
Definition: aacenc.h:84
g
const char * g
Definition: vf_curves.c:127
EIGHT_SHORT_SEQUENCE
@ EIGHT_SHORT_SEQUENCE
Definition: aac.h:65
s1
#define s1
Definition: regdef.h:38
LongTermPrediction::present
int8_t present
Definition: aacdec.h:75
ff_aac_encode_ltp_info
void ff_aac_encode_ltp_info(AACEncContext *s, SingleChannelElement *sce, int common_window)
Encode LTP data.
Definition: aacenc_ltp.c:35
IndividualChannelStream
Individual Channel Stream.
Definition: aacdec.h:84
NULL
#define NULL
Definition: coverity.c:32
aacenc_quantization.h
LongTermPrediction::coef_idx
int coef_idx
Definition: aacenc.h:69
FFPsyBand
single band psychoacoustic information
Definition: psymodel.h:50
IndividualChannelStream::predictor_present
int predictor_present
Definition: aacdec.h:95
SingleChannelElement::sf_idx
int sf_idx[128]
scalefactor indices
Definition: aacenc.h:119
ff_aac_ltp_insert_new_frame
void ff_aac_ltp_insert_new_frame(AACEncContext *s)
Definition: aacenc_ltp.c:53
IndividualChannelStream::window_sequence
enum WindowSequence window_sequence[2]
Definition: aacdec.h:86
quant_array_idx
static int quant_array_idx(const float val, const float *arr, const int num)
Definition: aacenc_utils.h:147
ChannelElement::common_window
int common_window
Set if channels share a common 'IndividualChannelStream' in bitstream.
Definition: aacenc.h:138
ff_aac_adjust_common_ltp
void ff_aac_adjust_common_ltp(AACEncContext *s, ChannelElement *cpe)
Definition: aacenc_ltp.c:130
SingleChannelElement::band_type
enum BandType band_type[128]
band types
Definition: aacdec.h:134
generate_samples
static void generate_samples(float *buf, LongTermPrediction *ltp)
Definition: aacenc_ltp.c:99
SingleChannelElement
Single Channel Element - used for both SCE and LFE elements.
Definition: aacdec.h:131
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:255
IndividualChannelStream::num_windows
int num_windows
Definition: aacdec.h:93
FFPsyBand::threshold
float threshold
Definition: psymodel.h:53
ChannelElement
channel element - generic struct for SCE/CPE/CCE/LFE
Definition: aacdec.h:148
FFMIN
#define FFMIN(a, b)
Definition: macros.h:49
LongTermPrediction::lag
int16_t lag
Definition: aacdec.h:76
tag
uint32_t tag
Definition: movenc.c:1786
AACEncContext
AAC encoder context.
Definition: aacenc.h:198
LongTermPrediction
Long Term Prediction.
Definition: aacdec.h:74
samples
Filter the word “frame” indicates either a video frame or a group of audio samples
Definition: filter_design.txt:8
quantize_band_cost
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)
Definition: aacenc_quantization.h:43
s0
#define s0
Definition: regdef.h:37
IndividualChannelStream::max_sfb
uint8_t max_sfb
number of scalefactor bands per group
Definition: aacdec.h:85
ff_aac_search_for_ltp
void ff_aac_search_for_ltp(AACEncContext *s, SingleChannelElement *sce, int common_window)
Mark LTP sfb's.
Definition: aacenc_ltp.c:159
SingleChannelElement::ltp_state
INTFLOAT ltp_state[3072]
time signal for LTP
Definition: aacdec.h:140
IndividualChannelStream::ltp
LongTermPrediction ltp
Definition: aacdec.h:90
aacenc_utils.h
IndividualChannelStream::group_len
uint8_t group_len[8]
Definition: aacdec.h:89
bits2
static const uint8_t bits2[81]
Definition: aactab.c:244