Line |
Branch |
Exec |
Source |
1 |
|
|
/* |
2 |
|
|
* Copyright (c) 2021 Paul B Mahol |
3 |
|
|
* |
4 |
|
|
* This file is part of FFmpeg. |
5 |
|
|
* |
6 |
|
|
* FFmpeg is free software; you can redistribute it and/or |
7 |
|
|
* modify it under the terms of the GNU Lesser General Public |
8 |
|
|
* License as published by the Free Software Foundation; either |
9 |
|
|
* version 2.1 of the License, or (at your option) any later version. |
10 |
|
|
* |
11 |
|
|
* FFmpeg is distributed in the hope that it will be useful, |
12 |
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 |
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 |
|
|
* Lesser General Public License for more details. |
15 |
|
|
* |
16 |
|
|
* You should have received a copy of the GNU Lesser General Public |
17 |
|
|
* License along with FFmpeg; if not, write to the Free Software |
18 |
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
19 |
|
|
*/ |
20 |
|
|
|
21 |
|
|
#include <float.h> |
22 |
|
|
|
23 |
|
|
#include "libavutil/channel_layout.h" |
24 |
|
|
#include "libavutil/common.h" |
25 |
|
|
#include "libavutil/mem.h" |
26 |
|
|
|
27 |
|
|
#include "avfilter.h" |
28 |
|
|
#include "filters.h" |
29 |
|
|
|
30 |
|
|
typedef struct ChanStats { |
31 |
|
|
double u; |
32 |
|
|
double v; |
33 |
|
|
double uv; |
34 |
|
|
} ChanStats; |
35 |
|
|
|
36 |
|
|
typedef struct AudioSDRContext { |
37 |
|
|
int channels; |
38 |
|
|
uint64_t nb_samples; |
39 |
|
|
double max; |
40 |
|
|
|
41 |
|
|
ChanStats *chs; |
42 |
|
|
|
43 |
|
|
AVFrame *cache[2]; |
44 |
|
|
|
45 |
|
|
int (*filter)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs); |
46 |
|
|
} AudioSDRContext; |
47 |
|
|
|
48 |
|
|
#define SDR_FILTER(name, type) \ |
49 |
|
|
static int sdr_##name(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)\ |
50 |
|
|
{ \ |
51 |
|
|
AudioSDRContext *s = ctx->priv; \ |
52 |
|
|
AVFrame *u = s->cache[0]; \ |
53 |
|
|
AVFrame *v = s->cache[1]; \ |
54 |
|
|
const int channels = u->ch_layout.nb_channels; \ |
55 |
|
|
const int start = (channels * jobnr) / nb_jobs; \ |
56 |
|
|
const int end = (channels * (jobnr+1)) / nb_jobs; \ |
57 |
|
|
const int nb_samples = u->nb_samples; \ |
58 |
|
|
\ |
59 |
|
|
for (int ch = start; ch < end; ch++) { \ |
60 |
|
|
ChanStats *chs = &s->chs[ch]; \ |
61 |
|
|
const type *const us = (type *)u->extended_data[ch]; \ |
62 |
|
|
const type *const vs = (type *)v->extended_data[ch]; \ |
63 |
|
|
double sum_uv = 0.; \ |
64 |
|
|
double sum_u = 0.; \ |
65 |
|
|
\ |
66 |
|
|
for (int n = 0; n < nb_samples; n++) { \ |
67 |
|
|
sum_u += us[n] * us[n]; \ |
68 |
|
|
sum_uv += (us[n] - vs[n]) * (us[n] - vs[n]); \ |
69 |
|
|
} \ |
70 |
|
|
\ |
71 |
|
|
chs->uv += sum_uv; \ |
72 |
|
|
chs->u += sum_u; \ |
73 |
|
|
} \ |
74 |
|
|
\ |
75 |
|
|
return 0; \ |
76 |
|
|
} |
77 |
|
|
|
78 |
|
✗ |
SDR_FILTER(fltp, float) |
79 |
|
✗ |
SDR_FILTER(dblp, double) |
80 |
|
|
|
81 |
|
|
#define SISDR_FILTER(name, type) \ |
82 |
|
|
static int sisdr_##name(AVFilterContext *ctx, void *arg,int jobnr,int nb_jobs)\ |
83 |
|
|
{ \ |
84 |
|
|
AudioSDRContext *s = ctx->priv; \ |
85 |
|
|
AVFrame *u = s->cache[0]; \ |
86 |
|
|
AVFrame *v = s->cache[1]; \ |
87 |
|
|
const int channels = u->ch_layout.nb_channels; \ |
88 |
|
|
const int start = (channels * jobnr) / nb_jobs; \ |
89 |
|
|
const int end = (channels * (jobnr+1)) / nb_jobs; \ |
90 |
|
|
const int nb_samples = u->nb_samples; \ |
91 |
|
|
\ |
92 |
|
|
for (int ch = start; ch < end; ch++) { \ |
93 |
|
|
ChanStats *chs = &s->chs[ch]; \ |
94 |
|
|
const type *const us = (type *)u->extended_data[ch]; \ |
95 |
|
|
const type *const vs = (type *)v->extended_data[ch]; \ |
96 |
|
|
double sum_uv = 0.; \ |
97 |
|
|
double sum_u = 0.; \ |
98 |
|
|
double sum_v = 0.; \ |
99 |
|
|
\ |
100 |
|
|
for (int n = 0; n < nb_samples; n++) { \ |
101 |
|
|
sum_u += us[n] * us[n]; \ |
102 |
|
|
sum_v += vs[n] * vs[n]; \ |
103 |
|
|
sum_uv += us[n] * vs[n]; \ |
104 |
|
|
} \ |
105 |
|
|
\ |
106 |
|
|
chs->uv += sum_uv; \ |
107 |
|
|
chs->u += sum_u; \ |
108 |
|
|
chs->v += sum_v; \ |
109 |
|
|
} \ |
110 |
|
|
\ |
111 |
|
|
return 0; \ |
112 |
|
|
} |
113 |
|
|
|
114 |
|
✗ |
SISDR_FILTER(fltp, float) |
115 |
|
✗ |
SISDR_FILTER(dblp, double) |
116 |
|
|
|
117 |
|
|
#define PSNR_FILTER(name, type) \ |
118 |
|
|
static int psnr_##name(AVFilterContext *ctx, void *arg, int jobnr,int nb_jobs)\ |
119 |
|
|
{ \ |
120 |
|
|
AudioSDRContext *s = ctx->priv; \ |
121 |
|
|
AVFrame *u = s->cache[0]; \ |
122 |
|
|
AVFrame *v = s->cache[1]; \ |
123 |
|
|
const int channels = u->ch_layout.nb_channels; \ |
124 |
|
|
const int start = (channels * jobnr) / nb_jobs; \ |
125 |
|
|
const int end = (channels * (jobnr+1)) / nb_jobs; \ |
126 |
|
|
const int nb_samples = u->nb_samples; \ |
127 |
|
|
\ |
128 |
|
|
for (int ch = start; ch < end; ch++) { \ |
129 |
|
|
ChanStats *chs = &s->chs[ch]; \ |
130 |
|
|
const type *const us = (type *)u->extended_data[ch]; \ |
131 |
|
|
const type *const vs = (type *)v->extended_data[ch]; \ |
132 |
|
|
double sum_uv = 0.; \ |
133 |
|
|
\ |
134 |
|
|
for (int n = 0; n < nb_samples; n++) \ |
135 |
|
|
sum_uv += (us[n] - vs[n]) * (us[n] - vs[n]); \ |
136 |
|
|
\ |
137 |
|
|
chs->uv += sum_uv; \ |
138 |
|
|
} \ |
139 |
|
|
\ |
140 |
|
|
return 0; \ |
141 |
|
|
} |
142 |
|
|
|
143 |
|
✗ |
PSNR_FILTER(fltp, float) |
144 |
|
✗ |
PSNR_FILTER(dblp, double) |
145 |
|
|
|
146 |
|
✗ |
static int activate(AVFilterContext *ctx) |
147 |
|
|
{ |
148 |
|
✗ |
AudioSDRContext *s = ctx->priv; |
149 |
|
✗ |
AVFilterLink *outlink = ctx->outputs[0]; |
150 |
|
|
int ret, status, available; |
151 |
|
|
int64_t pts; |
152 |
|
|
|
153 |
|
✗ |
FF_FILTER_FORWARD_STATUS_BACK_ALL(outlink, ctx); |
154 |
|
|
|
155 |
|
✗ |
available = FFMIN(ff_inlink_queued_samples(ctx->inputs[0]), ff_inlink_queued_samples(ctx->inputs[1])); |
156 |
|
✗ |
if (available > 0) { |
157 |
|
|
AVFrame *out; |
158 |
|
|
|
159 |
|
✗ |
for (int i = 0; i < 2; i++) { |
160 |
|
✗ |
ret = ff_inlink_consume_samples(ctx->inputs[i], available, available, &s->cache[i]); |
161 |
|
✗ |
if (ret < 0) { |
162 |
|
✗ |
av_frame_free(&s->cache[0]); |
163 |
|
✗ |
av_frame_free(&s->cache[1]); |
164 |
|
✗ |
return ret; |
165 |
|
|
} |
166 |
|
|
} |
167 |
|
|
|
168 |
|
✗ |
if (!ctx->is_disabled) |
169 |
|
✗ |
ff_filter_execute(ctx, s->filter, NULL, NULL, |
170 |
|
✗ |
FFMIN(outlink->ch_layout.nb_channels, ff_filter_get_nb_threads(ctx))); |
171 |
|
|
|
172 |
|
✗ |
av_frame_free(&s->cache[1]); |
173 |
|
✗ |
out = s->cache[0]; |
174 |
|
✗ |
s->cache[0] = NULL; |
175 |
|
|
|
176 |
|
✗ |
s->nb_samples += available; |
177 |
|
✗ |
return ff_filter_frame(outlink, out); |
178 |
|
|
} |
179 |
|
|
|
180 |
|
✗ |
for (int i = 0; i < 2; i++) { |
181 |
|
✗ |
if (ff_inlink_acknowledge_status(ctx->inputs[i], &status, &pts)) { |
182 |
|
✗ |
ff_outlink_set_status(outlink, status, pts); |
183 |
|
✗ |
return 0; |
184 |
|
|
} |
185 |
|
|
} |
186 |
|
|
|
187 |
|
✗ |
if (ff_outlink_frame_wanted(outlink)) { |
188 |
|
✗ |
for (int i = 0; i < 2; i++) { |
189 |
|
✗ |
if (s->cache[i] || ff_inlink_queued_samples(ctx->inputs[i]) > 0) |
190 |
|
✗ |
continue; |
191 |
|
✗ |
ff_inlink_request_frame(ctx->inputs[i]); |
192 |
|
✗ |
return 0; |
193 |
|
|
} |
194 |
|
|
} |
195 |
|
|
|
196 |
|
✗ |
return FFERROR_NOT_READY; |
197 |
|
|
} |
198 |
|
|
|
199 |
|
✗ |
static int config_output(AVFilterLink *outlink) |
200 |
|
|
{ |
201 |
|
✗ |
AVFilterContext *ctx = outlink->src; |
202 |
|
✗ |
AVFilterLink *inlink = ctx->inputs[0]; |
203 |
|
✗ |
AudioSDRContext *s = ctx->priv; |
204 |
|
|
|
205 |
|
✗ |
s->channels = inlink->ch_layout.nb_channels; |
206 |
|
|
|
207 |
|
✗ |
if (!strcmp(ctx->filter->name, "asdr")) |
208 |
|
✗ |
s->filter = inlink->format == AV_SAMPLE_FMT_FLTP ? sdr_fltp : sdr_dblp; |
209 |
|
✗ |
else if (!strcmp(ctx->filter->name, "asisdr")) |
210 |
|
✗ |
s->filter = inlink->format == AV_SAMPLE_FMT_FLTP ? sisdr_fltp : sisdr_dblp; |
211 |
|
|
else |
212 |
|
✗ |
s->filter = inlink->format == AV_SAMPLE_FMT_FLTP ? psnr_fltp : psnr_dblp; |
213 |
|
✗ |
s->max = inlink->format == AV_SAMPLE_FMT_FLTP ? FLT_MAX : DBL_MAX; |
214 |
|
|
|
215 |
|
✗ |
s->chs = av_calloc(outlink->ch_layout.nb_channels, sizeof(*s->chs)); |
216 |
|
✗ |
if (!s->chs) |
217 |
|
✗ |
return AVERROR(ENOMEM); |
218 |
|
|
|
219 |
|
✗ |
return 0; |
220 |
|
|
} |
221 |
|
|
|
222 |
|
✗ |
static av_cold void uninit(AVFilterContext *ctx) |
223 |
|
|
{ |
224 |
|
✗ |
AudioSDRContext *s = ctx->priv; |
225 |
|
|
|
226 |
|
✗ |
if (!strcmp(ctx->filter->name, "asdr")) { |
227 |
|
✗ |
for (int ch = 0; ch < s->channels; ch++) |
228 |
|
✗ |
av_log(ctx, AV_LOG_INFO, "SDR ch%d: %g dB\n", ch, 10. * log10(s->chs[ch].u / s->chs[ch].uv)); |
229 |
|
✗ |
} else if (!strcmp(ctx->filter->name, "asisdr")) { |
230 |
|
✗ |
for (int ch = 0; ch < s->channels; ch++) { |
231 |
|
✗ |
double scale = s->chs[ch].uv / s->chs[ch].v; |
232 |
|
✗ |
double sisdr = scale * scale * s->chs[ch].v / fmax(0., s->chs[ch].u + scale*scale*s->chs[ch].v - 2.0*scale*s->chs[ch].uv); |
233 |
|
|
|
234 |
|
✗ |
av_log(ctx, AV_LOG_INFO, "SI-SDR ch%d: %g dB\n", ch, 10. * log10(sisdr)); |
235 |
|
|
} |
236 |
|
|
} else { |
237 |
|
✗ |
for (int ch = 0; ch < s->channels; ch++) { |
238 |
|
✗ |
double psnr = s->chs[ch].uv > 0.0 ? 2.0 * log(s->max) - log(s->nb_samples / s->chs[ch].uv) : INFINITY; |
239 |
|
|
|
240 |
|
✗ |
av_log(ctx, AV_LOG_INFO, "PSNR ch%d: %g dB\n", ch, psnr); |
241 |
|
|
} |
242 |
|
|
} |
243 |
|
|
|
244 |
|
✗ |
av_frame_free(&s->cache[0]); |
245 |
|
✗ |
av_frame_free(&s->cache[1]); |
246 |
|
|
|
247 |
|
✗ |
av_freep(&s->chs); |
248 |
|
✗ |
} |
249 |
|
|
|
250 |
|
|
static const AVFilterPad inputs[] = { |
251 |
|
|
{ |
252 |
|
|
.name = "input0", |
253 |
|
|
.type = AVMEDIA_TYPE_AUDIO, |
254 |
|
|
}, |
255 |
|
|
{ |
256 |
|
|
.name = "input1", |
257 |
|
|
.type = AVMEDIA_TYPE_AUDIO, |
258 |
|
|
}, |
259 |
|
|
}; |
260 |
|
|
|
261 |
|
|
static const AVFilterPad outputs[] = { |
262 |
|
|
{ |
263 |
|
|
.name = "default", |
264 |
|
|
.type = AVMEDIA_TYPE_AUDIO, |
265 |
|
|
.config_props = config_output, |
266 |
|
|
}, |
267 |
|
|
}; |
268 |
|
|
|
269 |
|
|
const AVFilter ff_af_asdr = { |
270 |
|
|
.name = "asdr", |
271 |
|
|
.description = NULL_IF_CONFIG_SMALL("Measure Audio Signal-to-Distortion Ratio."), |
272 |
|
|
.priv_size = sizeof(AudioSDRContext), |
273 |
|
|
.activate = activate, |
274 |
|
|
.uninit = uninit, |
275 |
|
|
.flags = AVFILTER_FLAG_METADATA_ONLY | |
276 |
|
|
AVFILTER_FLAG_SLICE_THREADS | |
277 |
|
|
AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL, |
278 |
|
|
FILTER_INPUTS(inputs), |
279 |
|
|
FILTER_OUTPUTS(outputs), |
280 |
|
|
FILTER_SAMPLEFMTS(AV_SAMPLE_FMT_FLTP, |
281 |
|
|
AV_SAMPLE_FMT_DBLP), |
282 |
|
|
}; |
283 |
|
|
|
284 |
|
|
const AVFilter ff_af_apsnr = { |
285 |
|
|
.name = "apsnr", |
286 |
|
|
.description = NULL_IF_CONFIG_SMALL("Measure Audio Peak Signal-to-Noise Ratio."), |
287 |
|
|
.priv_size = sizeof(AudioSDRContext), |
288 |
|
|
.activate = activate, |
289 |
|
|
.uninit = uninit, |
290 |
|
|
.flags = AVFILTER_FLAG_METADATA_ONLY | |
291 |
|
|
AVFILTER_FLAG_SLICE_THREADS | |
292 |
|
|
AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL, |
293 |
|
|
FILTER_INPUTS(inputs), |
294 |
|
|
FILTER_OUTPUTS(outputs), |
295 |
|
|
FILTER_SAMPLEFMTS(AV_SAMPLE_FMT_FLTP, |
296 |
|
|
AV_SAMPLE_FMT_DBLP), |
297 |
|
|
}; |
298 |
|
|
|
299 |
|
|
const AVFilter ff_af_asisdr = { |
300 |
|
|
.name = "asisdr", |
301 |
|
|
.description = NULL_IF_CONFIG_SMALL("Measure Audio Scale-Invariant Signal-to-Distortion Ratio."), |
302 |
|
|
.priv_size = sizeof(AudioSDRContext), |
303 |
|
|
.activate = activate, |
304 |
|
|
.uninit = uninit, |
305 |
|
|
.flags = AVFILTER_FLAG_METADATA_ONLY | |
306 |
|
|
AVFILTER_FLAG_SLICE_THREADS | |
307 |
|
|
AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL, |
308 |
|
|
FILTER_INPUTS(inputs), |
309 |
|
|
FILTER_OUTPUTS(outputs), |
310 |
|
|
FILTER_SAMPLEFMTS(AV_SAMPLE_FMT_FLTP, |
311 |
|
|
AV_SAMPLE_FMT_DBLP), |
312 |
|
|
}; |
313 |
|
|
|