1 |
|
|
/* |
2 |
|
|
* AAC encoder TNS |
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 temporal noise shaping |
25 |
|
|
* @author Rostislav Pehlivanov ( atomnuker gmail com ) |
26 |
|
|
*/ |
27 |
|
|
|
28 |
|
|
#include "libavutil/libm.h" |
29 |
|
|
#include "aacenc.h" |
30 |
|
|
#include "aacenc_tns.h" |
31 |
|
|
#include "aactab.h" |
32 |
|
|
#include "aacenc_utils.h" |
33 |
|
|
#include "aacenc_quantization.h" |
34 |
|
|
|
35 |
|
|
/* Could be set to 3 to save an additional bit at the cost of little quality */ |
36 |
|
|
#define TNS_Q_BITS 4 |
37 |
|
|
|
38 |
|
|
/* Coefficient resolution in short windows */ |
39 |
|
|
#define TNS_Q_BITS_IS8 4 |
40 |
|
|
|
41 |
|
|
/* We really need the bits we save here elsewhere */ |
42 |
|
|
#define TNS_ENABLE_COEF_COMPRESSION |
43 |
|
|
|
44 |
|
|
/* TNS will only be used if the LPC gain is within these margins */ |
45 |
|
|
#define TNS_GAIN_THRESHOLD_LOW 1.4f |
46 |
|
|
#define TNS_GAIN_THRESHOLD_HIGH 1.16f*TNS_GAIN_THRESHOLD_LOW |
47 |
|
|
|
48 |
|
90 |
static inline int compress_coeffs(int *coef, int order, int c_bits) |
49 |
|
|
{ |
50 |
|
|
int i; |
51 |
✓✗ |
90 |
const int low_idx = c_bits ? 4 : 2; |
52 |
✓✗ |
90 |
const int shift_val = c_bits ? 8 : 4; |
53 |
✓✗ |
90 |
const int high_idx = c_bits ? 11 : 5; |
54 |
|
|
#ifndef TNS_ENABLE_COEF_COMPRESSION |
55 |
|
|
return 0; |
56 |
|
|
#endif /* TNS_ENABLE_COEF_COMPRESSION */ |
57 |
✓✓ |
674 |
for (i = 0; i < order; i++) |
58 |
✓✓✗✓
|
584 |
if (coef[i] >= low_idx && coef[i] <= high_idx) |
59 |
|
|
return 0; |
60 |
✓✓ |
674 |
for (i = 0; i < order; i++) |
61 |
✓✓ |
584 |
coef[i] -= (coef[i] > high_idx) ? shift_val : 0; |
62 |
|
90 |
return 1; |
63 |
|
|
} |
64 |
|
|
|
65 |
|
|
/** |
66 |
|
|
* Encode TNS data. |
67 |
|
|
* Coefficient compression is simply not lossless as it should be |
68 |
|
|
* on any decoder tested and as such is not active. |
69 |
|
|
*/ |
70 |
|
11743 |
void ff_aac_encode_tns_info(AACEncContext *s, SingleChannelElement *sce) |
71 |
|
|
{ |
72 |
|
11743 |
TemporalNoiseShaping *tns = &sce->tns; |
73 |
|
11743 |
int i, w, filt, coef_compress = 0, coef_len; |
74 |
|
11743 |
const int is8 = sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE; |
75 |
|
11743 |
const int c_bits = is8 ? TNS_Q_BITS_IS8 == 4 : TNS_Q_BITS == 4; |
76 |
|
|
|
77 |
✓✓ |
11743 |
if (!sce->tns.present) |
78 |
|
11697 |
return; |
79 |
|
|
|
80 |
✓✓ |
253 |
for (i = 0; i < sce->ics.num_windows; i++) { |
81 |
|
207 |
put_bits(&s->pb, 2 - is8, sce->tns.n_filt[i]); |
82 |
✓✓ |
207 |
if (!tns->n_filt[i]) |
83 |
|
140 |
continue; |
84 |
|
67 |
put_bits(&s->pb, 1, c_bits); |
85 |
✓✓ |
157 |
for (filt = 0; filt < tns->n_filt[i]; filt++) { |
86 |
|
90 |
put_bits(&s->pb, 6 - 2 * is8, tns->length[i][filt]); |
87 |
|
90 |
put_bits(&s->pb, 5 - 2 * is8, tns->order[i][filt]); |
88 |
✗✓ |
90 |
if (!tns->order[i][filt]) |
89 |
|
|
continue; |
90 |
|
90 |
put_bits(&s->pb, 1, tns->direction[i][filt]); |
91 |
|
90 |
coef_compress = compress_coeffs(tns->coef_idx[i][filt], |
92 |
|
|
tns->order[i][filt], c_bits); |
93 |
|
90 |
put_bits(&s->pb, 1, coef_compress); |
94 |
|
90 |
coef_len = c_bits + 3 - coef_compress; |
95 |
✓✓ |
674 |
for (w = 0; w < tns->order[i][filt]; w++) |
96 |
|
584 |
put_bits(&s->pb, coef_len, tns->coef_idx[i][filt][w]); |
97 |
|
|
} |
98 |
|
|
} |
99 |
|
|
} |
100 |
|
|
|
101 |
|
|
/* Apply TNS filter */ |
102 |
|
3515 |
void ff_aac_apply_tns(AACEncContext *s, SingleChannelElement *sce) |
103 |
|
|
{ |
104 |
|
3515 |
TemporalNoiseShaping *tns = &sce->tns; |
105 |
|
3515 |
IndividualChannelStream *ics = &sce->ics; |
106 |
|
|
int w, filt, m, i, top, order, bottom, start, end, size, inc; |
107 |
|
3515 |
const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb); |
108 |
|
|
float lpc[TNS_MAX_ORDER]; |
109 |
|
|
|
110 |
✓✓ |
7471 |
for (w = 0; w < ics->num_windows; w++) { |
111 |
|
3956 |
bottom = ics->num_swb; |
112 |
✓✓ |
4046 |
for (filt = 0; filt < tns->n_filt[w]; filt++) { |
113 |
|
90 |
top = bottom; |
114 |
|
90 |
bottom = FFMAX(0, top - tns->length[w][filt]); |
115 |
|
90 |
order = tns->order[w][filt]; |
116 |
✗✓ |
90 |
if (order == 0) |
117 |
|
|
continue; |
118 |
|
|
|
119 |
|
|
// tns_decode_coef |
120 |
|
90 |
compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0); |
121 |
|
|
|
122 |
✗✓ |
90 |
start = ics->swb_offset[FFMIN(bottom, mmm)]; |
123 |
✓✓ |
90 |
end = ics->swb_offset[FFMIN( top, mmm)]; |
124 |
✗✓ |
90 |
if ((size = end - start) <= 0) |
125 |
|
|
continue; |
126 |
✓✓ |
90 |
if (tns->direction[w][filt]) { |
127 |
|
41 |
inc = -1; |
128 |
|
41 |
start = end - 1; |
129 |
|
|
} else { |
130 |
|
49 |
inc = 1; |
131 |
|
|
} |
132 |
|
90 |
start += w * 128; |
133 |
|
|
|
134 |
|
|
/* AR filter */ |
135 |
✓✓ |
16290 |
for (m = 0; m < size; m++, start += inc) { |
136 |
✓✓ |
115506 |
for (i = 1; i <= FFMIN(m, order); i++) { |
137 |
|
99306 |
sce->coeffs[start] += lpc[i-1]*sce->pcoeffs[start - i*inc]; |
138 |
|
|
} |
139 |
|
|
} |
140 |
|
|
} |
141 |
|
|
} |
142 |
|
3515 |
} |
143 |
|
|
|
144 |
|
|
/* |
145 |
|
|
* c_bits - 1 if 4 bit coefficients, 0 if 3 bit coefficients |
146 |
|
|
*/ |
147 |
|
90 |
static inline void quantize_coefs(double *coef, int *idx, float *lpc, int order, |
148 |
|
|
int c_bits) |
149 |
|
|
{ |
150 |
|
|
int i; |
151 |
|
90 |
const float *quant_arr = tns_tmp2_map[c_bits]; |
152 |
✓✓ |
674 |
for (i = 0; i < order; i++) { |
153 |
✓✗ |
584 |
idx[i] = quant_array_idx(coef[i], quant_arr, c_bits ? 16 : 8); |
154 |
|
584 |
lpc[i] = quant_arr[idx[i]]; |
155 |
|
|
} |
156 |
|
90 |
} |
157 |
|
|
|
158 |
|
|
/* |
159 |
|
|
* 3 bits per coefficient with 8 short windows |
160 |
|
|
*/ |
161 |
|
3515 |
void ff_aac_search_for_tns(AACEncContext *s, SingleChannelElement *sce) |
162 |
|
|
{ |
163 |
|
3515 |
TemporalNoiseShaping *tns = &sce->tns; |
164 |
|
3515 |
int w, g, count = 0; |
165 |
|
|
double gain, coefs[MAX_LPC_ORDER]; |
166 |
|
3515 |
const int mmm = FFMIN(sce->ics.tns_max_bands, sce->ics.max_sfb); |
167 |
|
3515 |
const int is8 = sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE; |
168 |
|
3515 |
const int c_bits = is8 ? TNS_Q_BITS_IS8 == 4 : TNS_Q_BITS == 4; |
169 |
|
3515 |
const int sfb_start = av_clip(tns_min_sfb[is8][s->samplerate_index], 0, mmm); |
170 |
|
3515 |
const int sfb_end = av_clip(sce->ics.num_swb, 0, mmm); |
171 |
✓✓✓✗
|
3515 |
const int order = is8 ? 7 : s->profile == FF_PROFILE_AAC_LOW ? 12 : TNS_MAX_ORDER; |
172 |
✓✓ |
7013 |
const int slant = sce->ics.window_sequence[0] == LONG_STOP_SEQUENCE ? 1 : |
173 |
✓✓ |
3498 |
sce->ics.window_sequence[0] == LONG_START_SEQUENCE ? 0 : 2; |
174 |
|
3515 |
const int sfb_len = sfb_end - sfb_start; |
175 |
|
3515 |
const int coef_len = sce->ics.swb_offset[sfb_end] - sce->ics.swb_offset[sfb_start]; |
176 |
|
|
|
177 |
✓✓✗✓
|
3515 |
if (coef_len <= 0 || sfb_len <= 0) { |
178 |
|
115 |
sce->tns.present = 0; |
179 |
|
115 |
return; |
180 |
|
|
} |
181 |
|
|
|
182 |
✓✓ |
7164 |
for (w = 0; w < sce->ics.num_windows; w++) { |
183 |
|
3764 |
float en[2] = {0.0f, 0.0f}; |
184 |
|
3764 |
int oc_start = 0, os_start = 0; |
185 |
|
3764 |
int coef_start = sce->ics.swb_offset[sfb_start]; |
186 |
|
|
|
187 |
✓✓✓✓
|
94831 |
for (g = sfb_start; g < sce->ics.num_swb && g <= sfb_end; g++) { |
188 |
|
91067 |
FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[w*16+g]; |
189 |
✓✓ |
91067 |
if (g > sfb_start + (sfb_len/2)) |
190 |
|
45263 |
en[1] += band->energy; |
191 |
|
|
else |
192 |
|
45804 |
en[0] += band->energy; |
193 |
|
|
} |
194 |
|
|
|
195 |
|
|
/* LPC */ |
196 |
|
3764 |
gain = ff_lpc_calc_ref_coefs_f(&s->lpc, &sce->coeffs[w*128 + coef_start], |
197 |
|
|
coef_len, order, coefs); |
198 |
|
|
|
199 |
✓✗✓✓ ✓✓✓✓
|
3764 |
if (!order || !isfinite(gain) || gain < TNS_GAIN_THRESHOLD_LOW || gain > TNS_GAIN_THRESHOLD_HIGH) |
200 |
|
3697 |
continue; |
201 |
|
|
|
202 |
✓✓✓✗
|
67 |
tns->n_filt[w] = is8 ? 1 : order != TNS_MAX_ORDER ? 2 : 3; |
203 |
✓✓ |
157 |
for (g = 0; g < tns->n_filt[w]; g++) { |
204 |
✓✓ |
90 |
tns->direction[w][g] = slant != 2 ? slant : en[g] < en[!g]; |
205 |
✓✗ |
90 |
tns->order[w][g] = g < tns->n_filt[w] ? order/tns->n_filt[w] : order - oc_start; |
206 |
✓✗ |
90 |
tns->length[w][g] = g < tns->n_filt[w] ? sfb_len/tns->n_filt[w] : sfb_len - os_start; |
207 |
|
90 |
quantize_coefs(&coefs[oc_start], tns->coef_idx[w][g], tns->coef[w][g], |
208 |
|
|
tns->order[w][g], c_bits); |
209 |
|
90 |
oc_start += tns->order[w][g]; |
210 |
|
90 |
os_start += tns->length[w][g]; |
211 |
|
|
} |
212 |
|
67 |
count++; |
213 |
|
|
} |
214 |
|
3400 |
sce->tns.present = !!count; |
215 |
|
|
} |