1 /*
2 * Copyright (c) 2012 Clément Bœsch
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (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
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License along
17 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
19 */
20
21 /**
22 * @file
23 * EBU R.128 implementation
24 * @see http://tech.ebu.ch/loudness
25 * @see https://www.youtube.com/watch?v=iuEtQqC-Sqo "EBU R128 Introduction - Florian Camerer"
26 * @todo implement start/stop/reset through filter command injection
27 * @todo support other frequencies to avoid resampling
28 */
29
30 #include <math.h>
31
32 #include "libavutil/avassert.h"
33 #include "libavutil/avstring.h"
34 #include "libavutil/channel_layout.h"
35 #include "libavutil/dict.h"
36 #include "libavutil/xga_font_data.h"
37 #include "libavutil/opt.h"
38 #include "libavutil/timestamp.h"
39 #include "libswresample/swresample.h"
40 #include "audio.h"
41 #include "avfilter.h"
42 #include "formats.h"
43 #include "internal.h"
44
45 #define MAX_CHANNELS 63
46
47 /* pre-filter coefficients */
48 #define PRE_B0 1.53512485958697
49 #define PRE_B1 -2.69169618940638
50 #define PRE_B2 1.19839281085285
51 #define PRE_A1 -1.69065929318241
52 #define PRE_A2 0.73248077421585
53
54 /* RLB-filter coefficients */
55 #define RLB_B0 1.0
56 #define RLB_B1 -2.0
57 #define RLB_B2 1.0
58 #define RLB_A1 -1.99004745483398
59 #define RLB_A2 0.99007225036621
60
61 #define ABS_THRES -70 ///< silence gate: we discard anything below this absolute (LUFS) threshold
62 #define ABS_UP_THRES 10 ///< upper loud limit to consider (ABS_THRES being the minimum)
63 #define HIST_GRAIN 100 ///< defines histogram precision
64 #define HIST_SIZE ((ABS_UP_THRES - ABS_THRES) * HIST_GRAIN + 1)
65
66 /**
67 * A histogram is an array of HIST_SIZE hist_entry storing all the energies
68 * recorded (with an accuracy of 1/HIST_GRAIN) of the loudnesses from ABS_THRES
69 * (at 0) to ABS_UP_THRES (at HIST_SIZE-1).
70 * This fixed-size system avoids the need of a list of energies growing
71 * infinitely over the time and is thus more scalable.
72 */
73 struct hist_entry {
74 int count; ///< how many times the corresponding value occurred
75 double energy; ///< E = 10^((L + 0.691) / 10)
76 double loudness; ///< L = -0.691 + 10 * log10(E)
77 };
78
79 struct integrator {
80 double *cache[MAX_CHANNELS]; ///< window of filtered samples (N ms)
81 int cache_pos; ///< focus on the last added bin in the cache array
82 double sum[MAX_CHANNELS]; ///< sum of the last N ms filtered samples (cache content)
83 int filled; ///< 1 if the cache is completely filled, 0 otherwise
84 double rel_threshold; ///< relative threshold
85 double sum_kept_powers; ///< sum of the powers (weighted sums) above absolute threshold
86 int nb_kept_powers; ///< number of sum above absolute threshold
87 struct hist_entry *histogram; ///< histogram of the powers, used to compute LRA and I
88 };
89
90 struct rect { int x, y, w, h; };
91
92 typedef struct {
93 const AVClass *class; ///< AVClass context for log and options purpose
94
95 /* peak metering */
96 int peak_mode; ///< enabled peak modes
97 double *true_peaks; ///< true peaks per channel
98 double *sample_peaks; ///< sample peaks per channel
99 double *true_peaks_per_frame; ///< true peaks in a frame per channel
100 #if CONFIG_SWRESAMPLE
101 SwrContext *swr_ctx; ///< over-sampling context for true peak metering
102
103 /* type pun fix */
104 union {
105 double *t_double; ///< resampled audio data for true peak metering
106 uint8_t *t_uint8_t;
107 } swr_buf;
108
109 int swr_linesize;
110 #endif
111
112 /* video */
113 int do_video; ///< 1 if video output enabled, 0 otherwise
114 int w, h; ///< size of the video output
115 struct rect text; ///< rectangle for the LU legend on the left
116 struct rect graph; ///< rectangle for the main graph in the center
117 struct rect gauge; ///< rectangle for the gauge on the right
118 AVFrame *outpicref; ///< output picture reference, updated regularly
119 int meter; ///< select a EBU mode between +9 and +18
120 int scale_range; ///< the range of LU values according to the meter
121 int y_zero_lu; ///< the y value (pixel position) for 0 LU
122 int *y_line_ref; ///< y reference values for drawing the LU lines in the graph and the gauge
123
124 /* audio */
125 int nb_channels; ///< number of channels in the input
126 double *ch_weighting; ///< channel weighting mapping
127 int sample_count; ///< sample count used for refresh frequency, reset at refresh
128
129 /* Filter caches.
130 * The mult by 3 in the following is for X[i], X[i-1] and X[i-2] */
131 double x[MAX_CHANNELS * 3]; ///< 3 input samples cache for each channel
132 double y[MAX_CHANNELS * 3]; ///< 3 pre-filter samples cache for each channel
133 double z[MAX_CHANNELS * 3]; ///< 3 RLB-filter samples cache for each channel
134
135 #define I400_BINS (48000 * 4 / 10)
136 #define I3000_BINS (48000 * 3)
137 struct integrator i400; ///< 400ms integrator, used for Momentary loudness (M), and Integrated loudness (I)
138 struct integrator i3000; ///< 3s integrator, used for Short term loudness (S), and Loudness Range (LRA)
139
140 /* I and LRA specific */
141 double integrated_loudness; ///< integrated loudness in LUFS (I)
142 double loudness_range; ///< loudness range in LU (LRA)
143 double lra_low, lra_high; ///< low and high LRA values
144
145 /* misc */
146 int loglevel; ///< log level for frame logging
147 int metadata; ///< whether or not to inject loudness results in frames
148 } EBUR128Context;
149
150 enum {
151 PEAK_MODE_NONE = 0,
152 PEAK_MODE_SAMPLES_PEAKS = 1<<1,
153 PEAK_MODE_TRUE_PEAKS = 1<<2,
154 };
155
156 #define OFFSET(x) offsetof(EBUR128Context, x)
157 #define A AV_OPT_FLAG_AUDIO_PARAM
158 #define V AV_OPT_FLAG_VIDEO_PARAM
159 #define F AV_OPT_FLAG_FILTERING_PARAM
160 static const AVOption ebur128_options[] = {
161 { "video", "set video output", OFFSET(do_video), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, V|F },
162 { "size", "set video size", OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "640x480"}, 0, 0, V|F },
163 { "meter", "set scale meter (+9 to +18)", OFFSET(meter), AV_OPT_TYPE_INT, {.i64 = 9}, 9, 18, V|F },
164 { "framelog", "force frame logging level", OFFSET(loglevel), AV_OPT_TYPE_INT, {.i64 = -1}, INT_MIN, INT_MAX, A|V|F, "level" },
165 { "info", "information logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_INFO}, INT_MIN, INT_MAX, A|V|F, "level" },
166 { "verbose", "verbose logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_VERBOSE}, INT_MIN, INT_MAX, A|V|F, "level" },
167 { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, A|V|F },
168 { "peak", "set peak mode", OFFSET(peak_mode), AV_OPT_TYPE_FLAGS, {.i64 = PEAK_MODE_NONE}, 0, INT_MAX, A|F, "mode" },
169 { "none", "disable any peak mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_NONE}, INT_MIN, INT_MAX, A|F, "mode" },
170 { "sample", "enable peak-sample mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_SAMPLES_PEAKS}, INT_MIN, INT_MAX, A|F, "mode" },
171 { "true", "enable true-peak mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_TRUE_PEAKS}, INT_MIN, INT_MAX, A|F, "mode" },
172 { NULL },
173 };
174
175 AVFILTER_DEFINE_CLASS(ebur128);
176
177 static const uint8_t graph_colors[] = {
178 0xdd, 0x66, 0x66, // value above 0LU non reached
179 0x66, 0x66, 0xdd, // value below 0LU non reached
180 0x96, 0x33, 0x33, // value above 0LU reached
181 0x33, 0x33, 0x96, // value below 0LU reached
182 0xdd, 0x96, 0x96, // value above 0LU line non reached
183 0x96, 0x96, 0xdd, // value below 0LU line non reached
184 0xdd, 0x33, 0x33, // value above 0LU line reached
185 0x33, 0x33, 0xdd, // value below 0LU line reached
186 };
187
get_graph_color(const EBUR128Context * ebur128,int v,int y)188 static const uint8_t *get_graph_color(const EBUR128Context *ebur128, int v, int y)
189 {
190 const int below0 = y > ebur128->y_zero_lu;
191 const int reached = y >= v;
192 const int line = ebur128->y_line_ref[y] || y == ebur128->y_zero_lu;
193 const int colorid = 4*line + 2*reached + below0;
194 return graph_colors + 3*colorid;
195 }
196
lu_to_y(const EBUR128Context * ebur128,double v)197 static inline int lu_to_y(const EBUR128Context *ebur128, double v)
198 {
199 v += 2 * ebur128->meter; // make it in range [0;...]
200 v = av_clipf(v, 0, ebur128->scale_range); // make sure it's in the graph scale
201 v = ebur128->scale_range - v; // invert value (y=0 is on top)
202 return v * ebur128->graph.h / ebur128->scale_range; // rescale from scale range to px height
203 }
204
205 #define FONT8 0
206 #define FONT16 1
207
208 static const uint8_t font_colors[] = {
209 0xdd, 0xdd, 0x00,
210 0x00, 0x96, 0x96,
211 };
212
drawtext(AVFrame * pic,int x,int y,int ftid,const uint8_t * color,const char * fmt,...)213 static void drawtext(AVFrame *pic, int x, int y, int ftid, const uint8_t *color, const char *fmt, ...)
214 {
215 int i;
216 char buf[128] = {0};
217 const uint8_t *font;
218 int font_height;
219 va_list vl;
220
221 if (ftid == FONT16) font = avpriv_vga16_font, font_height = 16;
222 else if (ftid == FONT8) font = avpriv_cga_font, font_height = 8;
223 else return;
224
225 va_start(vl, fmt);
226 vsnprintf(buf, sizeof(buf), fmt, vl);
227 va_end(vl);
228
229 for (i = 0; buf[i]; i++) {
230 int char_y, mask;
231 uint8_t *p = pic->data[0] + y*pic->linesize[0] + (x + i*8)*3;
232
233 for (char_y = 0; char_y < font_height; char_y++) {
234 for (mask = 0x80; mask; mask >>= 1) {
235 if (font[buf[i] * font_height + char_y] & mask)
236 memcpy(p, color, 3);
237 else
238 memcpy(p, "\x00\x00\x00", 3);
239 p += 3;
240 }
241 p += pic->linesize[0] - 8*3;
242 }
243 }
244 }
245
drawline(AVFrame * pic,int x,int y,int len,int step)246 static void drawline(AVFrame *pic, int x, int y, int len, int step)
247 {
248 int i;
249 uint8_t *p = pic->data[0] + y*pic->linesize[0] + x*3;
250
251 for (i = 0; i < len; i++) {
252 memcpy(p, "\x00\xff\x00", 3);
253 p += step;
254 }
255 }
256
config_video_output(AVFilterLink * outlink)257 static int config_video_output(AVFilterLink *outlink)
258 {
259 int i, x, y;
260 uint8_t *p;
261 AVFilterContext *ctx = outlink->src;
262 EBUR128Context *ebur128 = ctx->priv;
263 AVFrame *outpicref;
264
265 /* check if there is enough space to represent everything decently */
266 if (ebur128->w < 640 || ebur128->h < 480) {
267 av_log(ctx, AV_LOG_ERROR, "Video size %dx%d is too small, "
268 "minimum size is 640x480\n", ebur128->w, ebur128->h);
269 return AVERROR(EINVAL);
270 }
271 outlink->w = ebur128->w;
272 outlink->h = ebur128->h;
273
274 #define PAD 8
275
276 /* configure text area position and size */
277 ebur128->text.x = PAD;
278 ebur128->text.y = 40;
279 ebur128->text.w = 3 * 8; // 3 characters
280 ebur128->text.h = ebur128->h - PAD - ebur128->text.y;
281
282 /* configure gauge position and size */
283 ebur128->gauge.w = 20;
284 ebur128->gauge.h = ebur128->text.h;
285 ebur128->gauge.x = ebur128->w - PAD - ebur128->gauge.w;
286 ebur128->gauge.y = ebur128->text.y;
287
288 /* configure graph position and size */
289 ebur128->graph.x = ebur128->text.x + ebur128->text.w + PAD;
290 ebur128->graph.y = ebur128->gauge.y;
291 ebur128->graph.w = ebur128->gauge.x - ebur128->graph.x - PAD;
292 ebur128->graph.h = ebur128->gauge.h;
293
294 /* graph and gauge share the LU-to-pixel code */
295 av_assert0(ebur128->graph.h == ebur128->gauge.h);
296
297 /* prepare the initial picref buffer */
298 av_frame_free(&ebur128->outpicref);
299 ebur128->outpicref = outpicref =
300 ff_get_video_buffer(outlink, outlink->w, outlink->h);
301 if (!outpicref)
302 return AVERROR(ENOMEM);
303 outlink->sample_aspect_ratio = (AVRational){1,1};
304
305 /* init y references values (to draw LU lines) */
306 ebur128->y_line_ref = av_calloc(ebur128->graph.h + 1, sizeof(*ebur128->y_line_ref));
307 if (!ebur128->y_line_ref)
308 return AVERROR(ENOMEM);
309
310 /* black background */
311 memset(outpicref->data[0], 0, ebur128->h * outpicref->linesize[0]);
312
313 /* draw LU legends */
314 drawtext(outpicref, PAD, PAD+16, FONT8, font_colors+3, " LU");
315 for (i = ebur128->meter; i >= -ebur128->meter * 2; i--) {
316 y = lu_to_y(ebur128, i);
317 x = PAD + (i < 10 && i > -10) * 8;
318 ebur128->y_line_ref[y] = i;
319 y -= 4; // -4 to center vertically
320 drawtext(outpicref, x, y + ebur128->graph.y, FONT8, font_colors+3,
321 "%c%d", i < 0 ? '-' : i > 0 ? '+' : ' ', FFABS(i));
322 }
323
324 /* draw graph */
325 ebur128->y_zero_lu = lu_to_y(ebur128, 0);
326 p = outpicref->data[0] + ebur128->graph.y * outpicref->linesize[0]
327 + ebur128->graph.x * 3;
328 for (y = 0; y < ebur128->graph.h; y++) {
329 const uint8_t *c = get_graph_color(ebur128, INT_MAX, y);
330
331 for (x = 0; x < ebur128->graph.w; x++)
332 memcpy(p + x*3, c, 3);
333 p += outpicref->linesize[0];
334 }
335
336 /* draw fancy rectangles around the graph and the gauge */
337 #define DRAW_RECT(r) do { \
338 drawline(outpicref, r.x, r.y - 1, r.w, 3); \
339 drawline(outpicref, r.x, r.y + r.h, r.w, 3); \
340 drawline(outpicref, r.x - 1, r.y, r.h, outpicref->linesize[0]); \
341 drawline(outpicref, r.x + r.w, r.y, r.h, outpicref->linesize[0]); \
342 } while (0)
343 DRAW_RECT(ebur128->graph);
344 DRAW_RECT(ebur128->gauge);
345
346 outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
347
348 return 0;
349 }
350
config_audio_input(AVFilterLink * inlink)351 static int config_audio_input(AVFilterLink *inlink)
352 {
353 AVFilterContext *ctx = inlink->dst;
354 EBUR128Context *ebur128 = ctx->priv;
355
356 /* Force 100ms framing in case of metadata injection: the frames must have
357 * a granularity of the window overlap to be accurately exploited.
358 * As for the true peaks mode, it just simplifies the resampling buffer
359 * allocation and the lookup in it (since sample buffers differ in size, it
360 * can be more complex to integrate in the one-sample loop of
361 * filter_frame()). */
362 if (ebur128->metadata || (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS))
363 inlink->min_samples =
364 inlink->max_samples =
365 inlink->partial_buf_size = inlink->sample_rate / 10;
366 return 0;
367 }
368
config_audio_output(AVFilterLink * outlink)369 static int config_audio_output(AVFilterLink *outlink)
370 {
371 int i;
372 AVFilterContext *ctx = outlink->src;
373 EBUR128Context *ebur128 = ctx->priv;
374 const int nb_channels = av_get_channel_layout_nb_channels(outlink->channel_layout);
375
376 #define BACK_MASK (AV_CH_BACK_LEFT |AV_CH_BACK_CENTER |AV_CH_BACK_RIGHT| \
377 AV_CH_TOP_BACK_LEFT|AV_CH_TOP_BACK_CENTER|AV_CH_TOP_BACK_RIGHT| \
378 AV_CH_SIDE_LEFT |AV_CH_SIDE_RIGHT| \
379 AV_CH_SURROUND_DIRECT_LEFT |AV_CH_SURROUND_DIRECT_RIGHT)
380
381 ebur128->nb_channels = nb_channels;
382 ebur128->ch_weighting = av_calloc(nb_channels, sizeof(*ebur128->ch_weighting));
383 if (!ebur128->ch_weighting)
384 return AVERROR(ENOMEM);
385
386 for (i = 0; i < nb_channels; i++) {
387 /* channel weighting */
388 const uint16_t chl = av_channel_layout_extract_channel(outlink->channel_layout, i);
389 if (chl & (AV_CH_LOW_FREQUENCY|AV_CH_LOW_FREQUENCY_2)) {
390 ebur128->ch_weighting[i] = 0;
391 } else if (chl & BACK_MASK) {
392 ebur128->ch_weighting[i] = 1.41;
393 } else {
394 ebur128->ch_weighting[i] = 1.0;
395 }
396
397 if (!ebur128->ch_weighting[i])
398 continue;
399
400 /* bins buffer for the two integration window (400ms and 3s) */
401 ebur128->i400.cache[i] = av_calloc(I400_BINS, sizeof(*ebur128->i400.cache[0]));
402 ebur128->i3000.cache[i] = av_calloc(I3000_BINS, sizeof(*ebur128->i3000.cache[0]));
403 if (!ebur128->i400.cache[i] || !ebur128->i3000.cache[i])
404 return AVERROR(ENOMEM);
405 }
406
407 outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
408
409 #if CONFIG_SWRESAMPLE
410 if (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS) {
411 int ret;
412
413 ebur128->swr_buf.t_double = av_malloc_array(nb_channels, 19200 * sizeof(double));
414 ebur128->true_peaks = av_calloc(nb_channels, sizeof(*ebur128->true_peaks));
415 ebur128->true_peaks_per_frame = av_calloc(nb_channels, sizeof(*ebur128->true_peaks_per_frame));
416 ebur128->swr_ctx = swr_alloc();
417 if (!ebur128->swr_buf.t_double || !ebur128->true_peaks ||
418 !ebur128->true_peaks_per_frame || !ebur128->swr_ctx)
419 return AVERROR(ENOMEM);
420
421 av_opt_set_int(ebur128->swr_ctx, "in_channel_layout", outlink->channel_layout, 0);
422 av_opt_set_int(ebur128->swr_ctx, "in_sample_rate", outlink->sample_rate, 0);
423 av_opt_set_sample_fmt(ebur128->swr_ctx, "in_sample_fmt", outlink->format, 0);
424
425 av_opt_set_int(ebur128->swr_ctx, "out_channel_layout", outlink->channel_layout, 0);
426 av_opt_set_int(ebur128->swr_ctx, "out_sample_rate", 192000, 0);
427 av_opt_set_sample_fmt(ebur128->swr_ctx, "out_sample_fmt", outlink->format, 0);
428
429 ret = swr_init(ebur128->swr_ctx);
430 if (ret < 0)
431 return ret;
432 }
433 #endif
434
435 if (ebur128->peak_mode & PEAK_MODE_SAMPLES_PEAKS) {
436 ebur128->sample_peaks = av_calloc(nb_channels, sizeof(*ebur128->sample_peaks));
437 if (!ebur128->sample_peaks)
438 return AVERROR(ENOMEM);
439 }
440
441 return 0;
442 }
443
444 #define ENERGY(loudness) (pow(10, ((loudness) + 0.691) / 10.))
445 #define LOUDNESS(energy) (-0.691 + 10 * log10(energy))
446 #define DBFS(energy) (20 * log10(energy))
447
get_histogram(void)448 static struct hist_entry *get_histogram(void)
449 {
450 int i;
451 struct hist_entry *h = av_calloc(HIST_SIZE, sizeof(*h));
452
453 if (!h)
454 return NULL;
455 for (i = 0; i < HIST_SIZE; i++) {
456 h[i].loudness = i / (double)HIST_GRAIN + ABS_THRES;
457 h[i].energy = ENERGY(h[i].loudness);
458 }
459 return h;
460 }
461
init(AVFilterContext * ctx)462 static av_cold int init(AVFilterContext *ctx)
463 {
464 EBUR128Context *ebur128 = ctx->priv;
465 AVFilterPad pad;
466
467 if (ebur128->loglevel != AV_LOG_INFO &&
468 ebur128->loglevel != AV_LOG_VERBOSE) {
469 if (ebur128->do_video || ebur128->metadata)
470 ebur128->loglevel = AV_LOG_VERBOSE;
471 else
472 ebur128->loglevel = AV_LOG_INFO;
473 }
474
475 if (!CONFIG_SWRESAMPLE && (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS)) {
476 av_log(ctx, AV_LOG_ERROR,
477 "True-peak mode requires libswresample to be performed\n");
478 return AVERROR(EINVAL);
479 }
480
481 // if meter is +9 scale, scale range is from -18 LU to +9 LU (or 3*9)
482 // if meter is +18 scale, scale range is from -36 LU to +18 LU (or 3*18)
483 ebur128->scale_range = 3 * ebur128->meter;
484
485 ebur128->i400.histogram = get_histogram();
486 ebur128->i3000.histogram = get_histogram();
487 if (!ebur128->i400.histogram || !ebur128->i3000.histogram)
488 return AVERROR(ENOMEM);
489
490 ebur128->integrated_loudness = ABS_THRES;
491 ebur128->loudness_range = 0;
492
493 /* insert output pads */
494 if (ebur128->do_video) {
495 pad = (AVFilterPad){
496 .name = av_strdup("out0"),
497 .type = AVMEDIA_TYPE_VIDEO,
498 .config_props = config_video_output,
499 };
500 if (!pad.name)
501 return AVERROR(ENOMEM);
502 ff_insert_outpad(ctx, 0, &pad);
503 }
504 pad = (AVFilterPad){
505 .name = av_asprintf("out%d", ebur128->do_video),
506 .type = AVMEDIA_TYPE_AUDIO,
507 .config_props = config_audio_output,
508 };
509 if (!pad.name)
510 return AVERROR(ENOMEM);
511 ff_insert_outpad(ctx, ebur128->do_video, &pad);
512
513 /* summary */
514 av_log(ctx, AV_LOG_VERBOSE, "EBU +%d scale\n", ebur128->meter);
515
516 return 0;
517 }
518
519 #define HIST_POS(power) (int)(((power) - ABS_THRES) * HIST_GRAIN)
520
521 /* loudness and power should be set such as loudness = -0.691 +
522 * 10*log10(power), we just avoid doing that calculus two times */
gate_update(struct integrator * integ,double power,double loudness,int gate_thres)523 static int gate_update(struct integrator *integ, double power,
524 double loudness, int gate_thres)
525 {
526 int ipower;
527 double relative_threshold;
528 int gate_hist_pos;
529
530 /* update powers histograms by incrementing current power count */
531 ipower = av_clip(HIST_POS(loudness), 0, HIST_SIZE - 1);
532 integ->histogram[ipower].count++;
533
534 /* compute relative threshold and get its position in the histogram */
535 integ->sum_kept_powers += power;
536 integ->nb_kept_powers++;
537 relative_threshold = integ->sum_kept_powers / integ->nb_kept_powers;
538 if (!relative_threshold)
539 relative_threshold = 1e-12;
540 integ->rel_threshold = LOUDNESS(relative_threshold) + gate_thres;
541 gate_hist_pos = av_clip(HIST_POS(integ->rel_threshold), 0, HIST_SIZE - 1);
542
543 return gate_hist_pos;
544 }
545
filter_frame(AVFilterLink * inlink,AVFrame * insamples)546 static int filter_frame(AVFilterLink *inlink, AVFrame *insamples)
547 {
548 int i, ch, idx_insample;
549 AVFilterContext *ctx = inlink->dst;
550 EBUR128Context *ebur128 = ctx->priv;
551 const int nb_channels = ebur128->nb_channels;
552 const int nb_samples = insamples->nb_samples;
553 const double *samples = (double *)insamples->data[0];
554 AVFrame *pic = ebur128->outpicref;
555
556 #if CONFIG_SWRESAMPLE
557 if (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS) {
558 const double *swr_samples = ebur128->swr_buf.t_double;
559 int ret = swr_convert(ebur128->swr_ctx, (uint8_t**)&ebur128->swr_buf, 19200,
560 (const uint8_t **)insamples->data, nb_samples);
561 if (ret < 0)
562 return ret;
563 for (ch = 0; ch < nb_channels; ch++)
564 ebur128->true_peaks_per_frame[ch] = 0.0;
565 for (idx_insample = 0; idx_insample < ret; idx_insample++) {
566 for (ch = 0; ch < nb_channels; ch++) {
567 ebur128->true_peaks[ch] = FFMAX(ebur128->true_peaks[ch], FFABS(*swr_samples));
568 ebur128->true_peaks_per_frame[ch] = FFMAX(ebur128->true_peaks_per_frame[ch],
569 FFABS(*swr_samples));
570 swr_samples++;
571 }
572 }
573 }
574 #endif
575
576 for (idx_insample = 0; idx_insample < nb_samples; idx_insample++) {
577 const int bin_id_400 = ebur128->i400.cache_pos;
578 const int bin_id_3000 = ebur128->i3000.cache_pos;
579
580 #define MOVE_TO_NEXT_CACHED_ENTRY(time) do { \
581 ebur128->i##time.cache_pos++; \
582 if (ebur128->i##time.cache_pos == I##time##_BINS) { \
583 ebur128->i##time.filled = 1; \
584 ebur128->i##time.cache_pos = 0; \
585 } \
586 } while (0)
587
588 MOVE_TO_NEXT_CACHED_ENTRY(400);
589 MOVE_TO_NEXT_CACHED_ENTRY(3000);
590
591 for (ch = 0; ch < nb_channels; ch++) {
592 double bin;
593
594 if (ebur128->peak_mode & PEAK_MODE_SAMPLES_PEAKS)
595 ebur128->sample_peaks[ch] = FFMAX(ebur128->sample_peaks[ch], FFABS(*samples));
596
597 ebur128->x[ch * 3] = *samples++; // set X[i]
598
599 if (!ebur128->ch_weighting[ch])
600 continue;
601
602 /* Y[i] = X[i]*b0 + X[i-1]*b1 + X[i-2]*b2 - Y[i-1]*a1 - Y[i-2]*a2 */
603 #define FILTER(Y, X, name) do { \
604 double *dst = ebur128->Y + ch*3; \
605 double *src = ebur128->X + ch*3; \
606 dst[2] = dst[1]; \
607 dst[1] = dst[0]; \
608 dst[0] = src[0]*name##_B0 + src[1]*name##_B1 + src[2]*name##_B2 \
609 - dst[1]*name##_A1 - dst[2]*name##_A2; \
610 } while (0)
611
612 // TODO: merge both filters in one?
613 FILTER(y, x, PRE); // apply pre-filter
614 ebur128->x[ch * 3 + 2] = ebur128->x[ch * 3 + 1];
615 ebur128->x[ch * 3 + 1] = ebur128->x[ch * 3 ];
616 FILTER(z, y, RLB); // apply RLB-filter
617
618 bin = ebur128->z[ch * 3] * ebur128->z[ch * 3];
619
620 /* add the new value, and limit the sum to the cache size (400ms or 3s)
621 * by removing the oldest one */
622 ebur128->i400.sum [ch] = ebur128->i400.sum [ch] + bin - ebur128->i400.cache [ch][bin_id_400];
623 ebur128->i3000.sum[ch] = ebur128->i3000.sum[ch] + bin - ebur128->i3000.cache[ch][bin_id_3000];
624
625 /* override old cache entry with the new value */
626 ebur128->i400.cache [ch][bin_id_400 ] = bin;
627 ebur128->i3000.cache[ch][bin_id_3000] = bin;
628 }
629
630 /* For integrated loudness, gating blocks are 400ms long with 75%
631 * overlap (see BS.1770-2 p5), so a re-computation is needed each 100ms
632 * (4800 samples at 48kHz). */
633 if (++ebur128->sample_count == 4800) {
634 double loudness_400, loudness_3000;
635 double power_400 = 1e-12, power_3000 = 1e-12;
636 AVFilterLink *outlink = ctx->outputs[0];
637 const int64_t pts = insamples->pts +
638 av_rescale_q(idx_insample, (AVRational){ 1, inlink->sample_rate },
639 outlink->time_base);
640 ebur128->sample_count = 0;
641
642 #define COMPUTE_LOUDNESS(m, time) do { \
643 if (ebur128->i##time.filled) { \
644 /* weighting sum of the last <time> ms */ \
645 for (ch = 0; ch < nb_channels; ch++) \
646 power_##time += ebur128->ch_weighting[ch] * ebur128->i##time.sum[ch]; \
647 power_##time /= I##time##_BINS; \
648 } \
649 loudness_##time = LOUDNESS(power_##time); \
650 } while (0)
651
652 COMPUTE_LOUDNESS(M, 400);
653 COMPUTE_LOUDNESS(S, 3000);
654
655 /* Integrated loudness */
656 #define I_GATE_THRES -10 // initially defined to -8 LU in the first EBU standard
657
658 if (loudness_400 >= ABS_THRES) {
659 double integrated_sum = 0;
660 int nb_integrated = 0;
661 int gate_hist_pos = gate_update(&ebur128->i400, power_400,
662 loudness_400, I_GATE_THRES);
663
664 /* compute integrated loudness by summing the histogram values
665 * above the relative threshold */
666 for (i = gate_hist_pos; i < HIST_SIZE; i++) {
667 const int nb_v = ebur128->i400.histogram[i].count;
668 nb_integrated += nb_v;
669 integrated_sum += nb_v * ebur128->i400.histogram[i].energy;
670 }
671 if (nb_integrated)
672 ebur128->integrated_loudness = LOUDNESS(integrated_sum / nb_integrated);
673 }
674
675 /* LRA */
676 #define LRA_GATE_THRES -20
677 #define LRA_LOWER_PRC 10
678 #define LRA_HIGHER_PRC 95
679
680 /* XXX: example code in EBU 3342 is ">=" but formula in BS.1770
681 * specs is ">" */
682 if (loudness_3000 >= ABS_THRES) {
683 int nb_powers = 0;
684 int gate_hist_pos = gate_update(&ebur128->i3000, power_3000,
685 loudness_3000, LRA_GATE_THRES);
686
687 for (i = gate_hist_pos; i < HIST_SIZE; i++)
688 nb_powers += ebur128->i3000.histogram[i].count;
689 if (nb_powers) {
690 int n, nb_pow;
691
692 /* get lower loudness to consider */
693 n = 0;
694 nb_pow = LRA_LOWER_PRC * nb_powers / 100. + 0.5;
695 for (i = gate_hist_pos; i < HIST_SIZE; i++) {
696 n += ebur128->i3000.histogram[i].count;
697 if (n >= nb_pow) {
698 ebur128->lra_low = ebur128->i3000.histogram[i].loudness;
699 break;
700 }
701 }
702
703 /* get higher loudness to consider */
704 n = nb_powers;
705 nb_pow = LRA_HIGHER_PRC * nb_powers / 100. + 0.5;
706 for (i = HIST_SIZE - 1; i >= 0; i--) {
707 n -= ebur128->i3000.histogram[i].count;
708 if (n < nb_pow) {
709 ebur128->lra_high = ebur128->i3000.histogram[i].loudness;
710 break;
711 }
712 }
713
714 // XXX: show low & high on the graph?
715 ebur128->loudness_range = ebur128->lra_high - ebur128->lra_low;
716 }
717 }
718
719 #define LOG_FMT "M:%6.1f S:%6.1f I:%6.1f LUFS LRA:%6.1f LU"
720
721 /* push one video frame */
722 if (ebur128->do_video) {
723 int x, y, ret;
724 uint8_t *p;
725
726 const int y_loudness_lu_graph = lu_to_y(ebur128, loudness_3000 + 23);
727 const int y_loudness_lu_gauge = lu_to_y(ebur128, loudness_400 + 23);
728
729 /* draw the graph using the short-term loudness */
730 p = pic->data[0] + ebur128->graph.y*pic->linesize[0] + ebur128->graph.x*3;
731 for (y = 0; y < ebur128->graph.h; y++) {
732 const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_graph, y);
733
734 memmove(p, p + 3, (ebur128->graph.w - 1) * 3);
735 memcpy(p + (ebur128->graph.w - 1) * 3, c, 3);
736 p += pic->linesize[0];
737 }
738
739 /* draw the gauge using the momentary loudness */
740 p = pic->data[0] + ebur128->gauge.y*pic->linesize[0] + ebur128->gauge.x*3;
741 for (y = 0; y < ebur128->gauge.h; y++) {
742 const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_gauge, y);
743
744 for (x = 0; x < ebur128->gauge.w; x++)
745 memcpy(p + x*3, c, 3);
746 p += pic->linesize[0];
747 }
748
749 /* draw textual info */
750 drawtext(pic, PAD, PAD - PAD/2, FONT16, font_colors,
751 LOG_FMT " ", // padding to erase trailing characters
752 loudness_400, loudness_3000,
753 ebur128->integrated_loudness, ebur128->loudness_range);
754
755 /* set pts and push frame */
756 pic->pts = pts;
757 ret = ff_filter_frame(outlink, av_frame_clone(pic));
758 if (ret < 0)
759 return ret;
760 }
761
762 if (ebur128->metadata) { /* happens only once per filter_frame call */
763 char metabuf[128];
764 #define META_PREFIX "lavfi.r128."
765
766 #define SET_META(name, var) do { \
767 snprintf(metabuf, sizeof(metabuf), "%.3f", var); \
768 av_dict_set(&insamples->metadata, name, metabuf, 0); \
769 } while (0)
770
771 #define SET_META_PEAK(name, ptype) do { \
772 if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
773 char key[64]; \
774 for (ch = 0; ch < nb_channels; ch++) { \
775 snprintf(key, sizeof(key), \
776 META_PREFIX AV_STRINGIFY(name) "_peaks_ch%d", ch); \
777 SET_META(key, ebur128->name##_peaks[ch]); \
778 } \
779 } \
780 } while (0)
781
782 SET_META(META_PREFIX "M", loudness_400);
783 SET_META(META_PREFIX "S", loudness_3000);
784 SET_META(META_PREFIX "I", ebur128->integrated_loudness);
785 SET_META(META_PREFIX "LRA", ebur128->loudness_range);
786 SET_META(META_PREFIX "LRA.low", ebur128->lra_low);
787 SET_META(META_PREFIX "LRA.high", ebur128->lra_high);
788
789 SET_META_PEAK(sample, SAMPLES);
790 SET_META_PEAK(true, TRUE);
791 }
792
793 av_log(ctx, ebur128->loglevel, "t: %-10s " LOG_FMT,
794 av_ts2timestr(pts, &outlink->time_base),
795 loudness_400, loudness_3000,
796 ebur128->integrated_loudness, ebur128->loudness_range);
797
798 #define PRINT_PEAKS(str, sp, ptype) do { \
799 if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
800 av_log(ctx, ebur128->loglevel, " " str ":"); \
801 for (ch = 0; ch < nb_channels; ch++) \
802 av_log(ctx, ebur128->loglevel, " %5.1f", DBFS(sp[ch])); \
803 av_log(ctx, ebur128->loglevel, " dBFS"); \
804 } \
805 } while (0)
806
807 PRINT_PEAKS("SPK", ebur128->sample_peaks, SAMPLES);
808 PRINT_PEAKS("FTPK", ebur128->true_peaks_per_frame, TRUE);
809 PRINT_PEAKS("TPK", ebur128->true_peaks, TRUE);
810 av_log(ctx, ebur128->loglevel, "\n");
811 }
812 }
813
814 return ff_filter_frame(ctx->outputs[ebur128->do_video], insamples);
815 }
816
query_formats(AVFilterContext * ctx)817 static int query_formats(AVFilterContext *ctx)
818 {
819 EBUR128Context *ebur128 = ctx->priv;
820 AVFilterFormats *formats;
821 AVFilterChannelLayouts *layouts;
822 AVFilterLink *inlink = ctx->inputs[0];
823 AVFilterLink *outlink = ctx->outputs[0];
824
825 static const enum AVSampleFormat sample_fmts[] = { AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_NONE };
826 static const int input_srate[] = {48000, -1}; // ITU-R BS.1770 provides coeff only for 48kHz
827 static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_RGB24, AV_PIX_FMT_NONE };
828
829 /* set optional output video format */
830 if (ebur128->do_video) {
831 formats = ff_make_format_list(pix_fmts);
832 if (!formats)
833 return AVERROR(ENOMEM);
834 ff_formats_ref(formats, &outlink->in_formats);
835 outlink = ctx->outputs[1];
836 }
837
838 /* set input and output audio formats
839 * Note: ff_set_common_* functions are not used because they affect all the
840 * links, and thus break the video format negotiation */
841 formats = ff_make_format_list(sample_fmts);
842 if (!formats)
843 return AVERROR(ENOMEM);
844 ff_formats_ref(formats, &inlink->out_formats);
845 ff_formats_ref(formats, &outlink->in_formats);
846
847 layouts = ff_all_channel_layouts();
848 if (!layouts)
849 return AVERROR(ENOMEM);
850 ff_channel_layouts_ref(layouts, &inlink->out_channel_layouts);
851 ff_channel_layouts_ref(layouts, &outlink->in_channel_layouts);
852
853 formats = ff_make_format_list(input_srate);
854 if (!formats)
855 return AVERROR(ENOMEM);
856 ff_formats_ref(formats, &inlink->out_samplerates);
857 ff_formats_ref(formats, &outlink->in_samplerates);
858
859 return 0;
860 }
861
uninit(AVFilterContext * ctx)862 static av_cold void uninit(AVFilterContext *ctx)
863 {
864 int i;
865 EBUR128Context *ebur128 = ctx->priv;
866
867 av_log(ctx, AV_LOG_INFO, "Summary:\n\n"
868 " Integrated loudness:\n"
869 " I: %5.1f LUFS\n"
870 " Threshold: %5.1f LUFS\n\n"
871 " Loudness range:\n"
872 " LRA: %5.1f LU\n"
873 " Threshold: %5.1f LUFS\n"
874 " LRA low: %5.1f LUFS\n"
875 " LRA high: %5.1f LUFS",
876 ebur128->integrated_loudness, ebur128->i400.rel_threshold,
877 ebur128->loudness_range, ebur128->i3000.rel_threshold,
878 ebur128->lra_low, ebur128->lra_high);
879
880 #define PRINT_PEAK_SUMMARY(str, sp, ptype) do { \
881 int ch; \
882 double maxpeak; \
883 maxpeak = 0.0; \
884 if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
885 for (ch = 0; ch < ebur128->nb_channels; ch++) \
886 maxpeak = FFMAX(maxpeak, sp[ch]); \
887 av_log(ctx, AV_LOG_INFO, "\n\n " str " peak:\n" \
888 " Peak: %5.1f dBFS", \
889 DBFS(maxpeak)); \
890 } \
891 } while (0)
892
893 PRINT_PEAK_SUMMARY("Sample", ebur128->sample_peaks, SAMPLES);
894 PRINT_PEAK_SUMMARY("True", ebur128->true_peaks, TRUE);
895 av_log(ctx, AV_LOG_INFO, "\n");
896
897 av_freep(&ebur128->y_line_ref);
898 av_freep(&ebur128->ch_weighting);
899 av_freep(&ebur128->true_peaks);
900 av_freep(&ebur128->sample_peaks);
901 av_freep(&ebur128->true_peaks_per_frame);
902 av_freep(&ebur128->i400.histogram);
903 av_freep(&ebur128->i3000.histogram);
904 for (i = 0; i < ebur128->nb_channels; i++) {
905 av_freep(&ebur128->i400.cache[i]);
906 av_freep(&ebur128->i3000.cache[i]);
907 }
908 for (i = 0; i < ctx->nb_outputs; i++)
909 av_freep(&ctx->output_pads[i].name);
910 av_frame_free(&ebur128->outpicref);
911 #if CONFIG_SWRESAMPLE
912 av_freep(&ebur128->swr_buf);
913 swr_free(&ebur128->swr_ctx);
914 #endif
915 }
916
917 static const AVFilterPad ebur128_inputs[] = {
918 {
919 .name = "default",
920 .type = AVMEDIA_TYPE_AUDIO,
921 .filter_frame = filter_frame,
922 .config_props = config_audio_input,
923 },
924 { NULL }
925 };
926
927 AVFilter ff_af_ebur128 = {
928 .name = "ebur128",
929 .description = NULL_IF_CONFIG_SMALL("EBU R128 scanner."),
930 .priv_size = sizeof(EBUR128Context),
931 .init = init,
932 .uninit = uninit,
933 .query_formats = query_formats,
934 .inputs = ebur128_inputs,
935 .outputs = NULL,
936 .priv_class = &ebur128_class,
937 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
938 };
939