1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
|
/*
* AAC encoder TNS
* Copyright (C) 2015 Rostislav Pehlivanov
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* AAC encoder temporal noise shaping
* @author Rostislav Pehlivanov ( atomnuker gmail com )
*/
#include "aacenc.h"
#include "aacenc_tns.h"
#include "aactab.h"
#include "aacenc_utils.h"
#include "aacenc_quantization.h"
static inline void conv_to_int32(int32_t *loc, float *samples, int num, float norm)
{
int i;
for (i = 0; i < num; i++)
loc[i] = ceilf((samples[i]/norm)*INT32_MAX);
}
static inline void conv_to_float(float *arr, int32_t *cof, int num)
{
int i;
for (i = 0; i < num; i++)
arr[i] = (float)cof[i]/INT32_MAX;
}
/* Input: quantized 4 bit coef, output: 1 if first (MSB) 2 bits are the same */
static inline int coef_test_compression(int coef)
{
int tmp = coef >> 2;
int res = ff_ctz(tmp);
if (res > 1)
return 1; /* ...00 -> compressable */
else if (res == 1)
return 0; /* ...10 -> uncompressable */
else if (ff_ctz(tmp >> 1) > 0)
return 0; /* ...0 1 -> uncompressable */
else
return 1; /* ...1 1 -> compressable */
}
static inline int compress_coef(int *coefs, int num)
{
int i, res = 0;
for (i = 0; i < num; i++)
res += coef_test_compression(coefs[i]);
return res == num ? 1 : 0;
}
/**
* Encode TNS data.
* Coefficient compression saves a single bit.
*/
void ff_aac_encode_tns_info(AACEncContext *s, SingleChannelElement *sce)
{
int i, w, filt, coef_len, coef_compress;
const int coef_res = MAX_LPC_PRECISION == 4 ? 1 : 0;
const int is8 = sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE;
put_bits(&s->pb, 1, !!sce->tns.present);
if (!sce->tns.present)
return;
for (i = 0; i < sce->ics.num_windows; i++) {
put_bits(&s->pb, 2 - is8, sce->tns.n_filt[i]);
if (sce->tns.n_filt[i]) {
put_bits(&s->pb, 1, !!coef_res);
for (filt = 0; filt < sce->tns.n_filt[i]; filt++) {
put_bits(&s->pb, 6 - 2 * is8, sce->tns.length[i][filt]);
put_bits(&s->pb, 5 - 2 * is8, sce->tns.order[i][filt]);
if (sce->tns.order[i][filt]) {
coef_compress = compress_coef(sce->tns.coef_idx[i][filt],
sce->tns.order[i][filt]);
put_bits(&s->pb, 1, !!sce->tns.direction[i][filt]);
put_bits(&s->pb, 1, !!coef_compress);
coef_len = coef_res + 3 - coef_compress;
for (w = 0; w < sce->tns.order[i][filt]; w++)
put_bits(&s->pb, coef_len, sce->tns.coef_idx[i][filt][w]);
}
}
}
}
}
static int process_tns_coeffs(TemporalNoiseShaping *tns, float *tns_coefs_raw,
int order, int w, int filt)
{
int i, j;
int *idx = tns->coef_idx[w][filt];
float *lpc = tns->coef[w][filt];
const int iqfac_p = ((1 << (MAX_LPC_PRECISION-1)) - 0.5)/(M_PI/2.0);
const int iqfac_m = ((1 << (MAX_LPC_PRECISION-1)) + 0.5)/(M_PI/2.0);
float temp[TNS_MAX_ORDER] = {0.0f}, out[TNS_MAX_ORDER] = {0.0f};
/* Quantization */
for (i = 0; i < order; i++) {
idx[i] = ceilf(asin(tns_coefs_raw[i])*((tns_coefs_raw[i] >= 0) ? iqfac_p : iqfac_m));
lpc[i] = 2*sin(idx[i]/((idx[i] >= 0) ? iqfac_p : iqfac_m));
}
/* Trim any coeff less than 0.1f from the end */
for (i = order; i > -1; i--) {
lpc[i] = (fabs(lpc[i]) > 0.1f) ? lpc[i] : 0.0f;
if (lpc[i] != 0.0 ) {
order = i;
break;
}
}
if (!order)
return 0;
/* Step up procedure, convert to LPC coeffs */
out[0] = 1.0f;
for (i = 1; i <= order; i++) {
for (j = 1; j < i; j++) {
temp[j] = out[j] + lpc[i]*out[i-j];
}
for (j = 1; j <= i; j++) {
out[j] = temp[j];
}
out[i] = lpc[i-1];
}
memcpy(lpc, out, TNS_MAX_ORDER*sizeof(float));
return order;
}
static void apply_tns_filter(float *out, float *in, int order, int direction,
float *tns_coefs, int ltp_used, int w, int filt,
int start_i, int len)
{
int i, j, inc, start = start_i;
float tmp[TNS_MAX_ORDER+1];
if (direction) {
inc = -1;
start = (start + len) - 1;
} else {
inc = 1;
}
if (!ltp_used) { /* AR filter */
for (i = 0; i < len; i++, start += inc)
out[i] = in[start];
for (j = 1; j <= FFMIN(i, order); j++)
out[i] += tns_coefs[j]*in[start - j*inc];
} else { /* MA filter */
for (i = 0; i < len; i++, start += inc) {
tmp[0] = out[i] = in[start];
for (j = 1; j <= FFMIN(i, order); j++)
out[i] += tmp[j]*tns_coefs[j];
for (j = order; j > 0; j--)
tmp[j] = tmp[j - 1];
}
}
}
void ff_aac_search_for_tns(AACEncContext *s, SingleChannelElement *sce)
{
TemporalNoiseShaping *tns = &sce->tns;
int w, g, order, sfb_start, sfb_len, coef_start, shift[MAX_LPC_ORDER], count = 0;
const int is8 = sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE;
const int tns_max_order = is8 ? 7 : s->profile == FF_PROFILE_AAC_LOW ? 12 : TNS_MAX_ORDER;
const float freq_mult = mpeg4audio_sample_rates[s->samplerate_index]/(1024.0f/sce->ics.num_windows)/2.0f;
float max_coef = 0.0f;
sce->tns.present = 0;
return;
for (coef_start = 0; coef_start < 1024; coef_start++)
max_coef = FFMAX(max_coef, sce->pcoeffs[coef_start]);
for (w = 0; w < sce->ics.num_windows; w++) {
int filters = 1, start = 0, coef_len = 0;
int32_t conv_coeff[1024] = {0};
int32_t coefs_t[MAX_LPC_ORDER][MAX_LPC_ORDER] = {{0}};
/* Determine start sfb + coef - excludes anything below threshold */
for (g = 0; g < sce->ics.num_swb; g++) {
if (start*freq_mult > TNS_LOW_LIMIT) {
sfb_start = w*16+g;
sfb_len = (w+1)*16 + g - sfb_start;
coef_start = sce->ics.swb_offset[sfb_start];
coef_len = sce->ics.swb_offset[sfb_start + sfb_len] - coef_start;
break;
}
start += sce->ics.swb_sizes[g];
}
if (coef_len <= 0)
continue;
conv_to_int32(conv_coeff, &sce->pcoeffs[coef_start], coef_len, max_coef);
/* LPC */
order = ff_lpc_calc_coefs(&s->lpc, conv_coeff, coef_len,
TNS_MIN_PRED_ORDER, tns_max_order,
32, coefs_t, shift,
FF_LPC_TYPE_LEVINSON, 10,
ORDER_METHOD_EST, MAX_LPC_SHIFT, 0) - 1;
/* Works surprisingly well, remember to tweak MAX_LPC_SHIFT if you want to play around with this */
if (shift[order] > 3) {
int direction = 0;
float tns_coefs_raw[TNS_MAX_ORDER];
tns->n_filt[w] = filters++;
conv_to_float(tns_coefs_raw, coefs_t[order], order);
for (g = 0; g < tns->n_filt[w]; g++) {
process_tns_coeffs(tns, tns_coefs_raw, order, w, g);
apply_tns_filter(&sce->coeffs[coef_start], sce->pcoeffs, order, direction, tns->coef[w][g],
sce->ics.ltp.present, w, g, coef_start, coef_len);
tns->order[w][g] = order;
tns->length[w][g] = sfb_len;
tns->direction[w][g] = direction;
}
count++;
}
}
sce->tns.present = !!count;
}
|
