/src/ffmpeg/libavcodec/aacenc_is.c
Line | Count | Source |
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 | | |
33 | | /** Frequency in Hz for lower limit of intensity stereo **/ |
34 | 0 | #define INT_STEREO_LOW_LIMIT 6100 |
35 | | |
36 | | struct AACISError { |
37 | | int pass; /* 1 if dist2 <= dist1 */ |
38 | | int phase; /* -1 or +1 */ |
39 | | float error; /* fabs(dist1 - dist2) */ |
40 | | float dist1; /* From original coeffs */ |
41 | | float dist2; /* From IS'd coeffs */ |
42 | | float ener01; |
43 | | }; |
44 | | |
45 | | static struct AACISError aac_is_encoding_err(AACEncContext *s, ChannelElement *cpe, |
46 | | int start, int w, int g, float ener0, |
47 | | float ener1, float ener01, int phase) |
48 | 0 | { |
49 | 0 | int i, w2; |
50 | 0 | SingleChannelElement *sce0 = &cpe->ch[0]; |
51 | 0 | SingleChannelElement *sce1 = &cpe->ch[1]; |
52 | 0 | float *L = sce0->coeffs; |
53 | 0 | float *R = sce1->coeffs; |
54 | 0 | float *L34 = &s->scoefs[256*0], *R34 = &s->scoefs[256*1]; |
55 | 0 | float *IS = &s->scoefs[256*2], *I34 = &s->scoefs[256*3]; |
56 | 0 | float dist1 = 0.0f, dist2 = 0.0f; |
57 | 0 | struct AACISError is_error = {0}; |
58 | |
|
59 | 0 | if (ener01 <= 0 || ener0 <= 0) { |
60 | 0 | is_error.pass = 0; |
61 | 0 | return is_error; |
62 | 0 | } |
63 | | |
64 | 0 | for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) { |
65 | 0 | FFPsyBand *band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)*16+g]; |
66 | 0 | FFPsyBand *band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)*16+g]; |
67 | 0 | int is_band_type, is_sf_idx = FFMAX(1, sce0->sf_idx[w*16+g]-4); |
68 | 0 | float e01_34 = phase*pos_pow34(ener1/ener0); |
69 | 0 | float maxval, dist_spec_err = 0.0f; |
70 | 0 | float minthr = FFMIN(band0->threshold, band1->threshold); |
71 | 0 | for (i = 0; i < sce0->ics.swb_sizes[g]; i++) |
72 | 0 | IS[i] = (L[start+(w+w2)*128+i] + phase*R[start+(w+w2)*128+i])*sqrt(ener0/ener01); |
73 | 0 | s->aacdsp.abs_pow34(L34, &L[start+(w+w2)*128], sce0->ics.swb_sizes[g]); |
74 | 0 | s->aacdsp.abs_pow34(R34, &R[start+(w+w2)*128], sce0->ics.swb_sizes[g]); |
75 | 0 | s->aacdsp.abs_pow34(I34, IS, sce0->ics.swb_sizes[g]); |
76 | 0 | maxval = find_max_val(1, sce0->ics.swb_sizes[g], I34); |
77 | 0 | is_band_type = find_min_book(maxval, is_sf_idx); |
78 | 0 | dist1 += quantize_band_cost(s, &L[start + (w+w2)*128], L34, |
79 | 0 | sce0->ics.swb_sizes[g], |
80 | 0 | sce0->sf_idx[w*16+g], |
81 | 0 | sce0->band_type[w*16+g], |
82 | 0 | s->lambda / band0->threshold, INFINITY, NULL, NULL); |
83 | 0 | dist1 += quantize_band_cost(s, &R[start + (w+w2)*128], R34, |
84 | 0 | sce1->ics.swb_sizes[g], |
85 | 0 | sce1->sf_idx[w*16+g], |
86 | 0 | sce1->band_type[w*16+g], |
87 | 0 | s->lambda / band1->threshold, INFINITY, NULL, NULL); |
88 | 0 | dist2 += quantize_band_cost(s, IS, I34, sce0->ics.swb_sizes[g], |
89 | 0 | is_sf_idx, is_band_type, |
90 | 0 | s->lambda / minthr, INFINITY, NULL, NULL); |
91 | 0 | for (i = 0; i < sce0->ics.swb_sizes[g]; i++) { |
92 | 0 | dist_spec_err += (L34[i] - I34[i])*(L34[i] - I34[i]); |
93 | 0 | dist_spec_err += (R34[i] - I34[i]*e01_34)*(R34[i] - I34[i]*e01_34); |
94 | 0 | } |
95 | 0 | dist_spec_err *= s->lambda / minthr; |
96 | 0 | dist2 += dist_spec_err; |
97 | 0 | } |
98 | |
|
99 | 0 | is_error.pass = dist2 <= dist1; |
100 | 0 | is_error.phase = phase; |
101 | 0 | is_error.error = dist2 - dist1; |
102 | 0 | is_error.dist1 = dist1; |
103 | 0 | is_error.dist2 = dist2; |
104 | 0 | is_error.ener01 = ener01; |
105 | |
|
106 | 0 | return is_error; |
107 | 0 | } |
108 | | |
109 | | void ff_aac_search_for_is(AACEncContext *s, AVCodecContext *avctx, ChannelElement *cpe) |
110 | 0 | { |
111 | 0 | SingleChannelElement *sce0 = &cpe->ch[0]; |
112 | 0 | SingleChannelElement *sce1 = &cpe->ch[1]; |
113 | 0 | int start = 0, count = 0, w, w2, g, i, prev_sf1 = -1, prev_bt = -1, prev_is = 0; |
114 | 0 | const float freq_mult = avctx->sample_rate/(1024.0f/sce0->ics.num_windows)/2.0f; |
115 | 0 | uint8_t nextband1[128]; |
116 | |
|
117 | 0 | if (!cpe->common_window) |
118 | 0 | return; |
119 | | |
120 | | /** Scout out next nonzero bands */ |
121 | 0 | ff_init_nextband_map(sce1, nextband1); |
122 | |
|
123 | 0 | for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) { |
124 | 0 | start = 0; |
125 | 0 | for (g = 0; g < sce0->ics.num_swb; g++) { |
126 | 0 | if (start*freq_mult > INT_STEREO_LOW_LIMIT*(s->lambda/170.0f) && |
127 | 0 | cpe->ch[0].band_type[w*16+g] != NOISE_BT && !cpe->ch[0].zeroes[w*16+g] && |
128 | 0 | cpe->ch[1].band_type[w*16+g] != NOISE_BT && !cpe->ch[1].zeroes[w*16+g] && |
129 | 0 | ff_sfdelta_can_remove_band(sce1, nextband1, prev_sf1, w*16+g)) { |
130 | 0 | float ener0 = 0.0f, ener1 = 0.0f, ener01 = 0.0f, ener01p = 0.0f; |
131 | 0 | struct AACISError ph_err1, ph_err2, *best; |
132 | 0 | for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) { |
133 | 0 | for (i = 0; i < sce0->ics.swb_sizes[g]; i++) { |
134 | 0 | float coef0 = sce0->coeffs[start+(w+w2)*128+i]; |
135 | 0 | float coef1 = sce1->coeffs[start+(w+w2)*128+i]; |
136 | 0 | ener0 += coef0*coef0; |
137 | 0 | ener1 += coef1*coef1; |
138 | 0 | ener01 += (coef0 + coef1)*(coef0 + coef1); |
139 | 0 | ener01p += (coef0 - coef1)*(coef0 - coef1); |
140 | 0 | } |
141 | 0 | } |
142 | 0 | ph_err1 = aac_is_encoding_err(s, cpe, start, w, g, |
143 | 0 | ener0, ener1, ener01p, -1); |
144 | 0 | ph_err2 = aac_is_encoding_err(s, cpe, start, w, g, |
145 | 0 | ener0, ener1, ener01, +1); |
146 | 0 | best = (ph_err1.pass && ph_err1.error < ph_err2.error) ? &ph_err1 : &ph_err2; |
147 | 0 | if (best->pass) { |
148 | 0 | cpe->is_mask[w*16+g] = 1; |
149 | 0 | cpe->ms_mask[w*16+g] = 0; |
150 | 0 | cpe->ch[0].is_ener[w*16+g] = sqrt(ener0 / best->ener01); |
151 | 0 | cpe->ch[1].is_ener[w*16+g] = ener0/ener1; |
152 | 0 | cpe->ch[1].band_type[w*16+g] = (best->phase > 0) ? INTENSITY_BT : INTENSITY_BT2; |
153 | 0 | if (prev_is && prev_bt != cpe->ch[1].band_type[w*16+g]) { |
154 | | /** Flip M/S mask and pick the other CB, since it encodes more efficiently */ |
155 | 0 | cpe->ms_mask[w*16+g] = 1; |
156 | 0 | cpe->ch[1].band_type[w*16+g] = (best->phase > 0) ? INTENSITY_BT2 : INTENSITY_BT; |
157 | 0 | } |
158 | 0 | prev_bt = cpe->ch[1].band_type[w*16+g]; |
159 | 0 | count++; |
160 | 0 | } |
161 | 0 | } |
162 | 0 | if (!sce1->zeroes[w*16+g] && sce1->band_type[w*16+g] < RESERVED_BT) |
163 | 0 | prev_sf1 = sce1->sf_idx[w*16+g]; |
164 | 0 | prev_is = cpe->is_mask[w*16+g]; |
165 | 0 | start += sce0->ics.swb_sizes[g]; |
166 | 0 | } |
167 | 0 | } |
168 | 0 | cpe->is_mode = !!count; |
169 | 0 | } |