FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
f_ebur128.c
Go to the documentation of this file.
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 True Peak
27  * @todo implement start/stop/reset through filter command injection
28  * @todo support other frequencies to avoid resampling
29  */
30 
31 #include <math.h>
32 
33 #include "libavutil/avassert.h"
34 #include "libavutil/avstring.h"
36 #include "libavutil/dict.h"
38 #include "libavutil/opt.h"
39 #include "libavutil/timestamp.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  /* video */
96  int do_video; ///< 1 if video output enabled, 0 otherwise
97  int w, h; ///< size of the video output
98  struct rect text; ///< rectangle for the LU legend on the left
99  struct rect graph; ///< rectangle for the main graph in the center
100  struct rect gauge; ///< rectangle for the gauge on the right
101  AVFrame *outpicref; ///< output picture reference, updated regularly
102  int meter; ///< select a EBU mode between +9 and +18
103  int scale_range; ///< the range of LU values according to the meter
104  int y_zero_lu; ///< the y value (pixel position) for 0 LU
105  int *y_line_ref; ///< y reference values for drawing the LU lines in the graph and the gauge
106 
107  /* audio */
108  int nb_channels; ///< number of channels in the input
109  double *ch_weighting; ///< channel weighting mapping
110  int sample_count; ///< sample count used for refresh frequency, reset at refresh
111 
112  /* Filter caches.
113  * The mult by 3 in the following is for X[i], X[i-1] and X[i-2] */
114  double x[MAX_CHANNELS * 3]; ///< 3 input samples cache for each channel
115  double y[MAX_CHANNELS * 3]; ///< 3 pre-filter samples cache for each channel
116  double z[MAX_CHANNELS * 3]; ///< 3 RLB-filter samples cache for each channel
117 
118 #define I400_BINS (48000 * 4 / 10)
119 #define I3000_BINS (48000 * 3)
120  struct integrator i400; ///< 400ms integrator, used for Momentary loudness (M), and Integrated loudness (I)
121  struct integrator i3000; ///< 3s integrator, used for Short term loudness (S), and Loudness Range (LRA)
122 
123  /* I and LRA specific */
124  double integrated_loudness; ///< integrated loudness in LUFS (I)
125  double loudness_range; ///< loudness range in LU (LRA)
126  double lra_low, lra_high; ///< low and high LRA values
127 
128  /* misc */
129  int loglevel; ///< log level for frame logging
130  int metadata; ///< whether or not to inject loudness results in frames
132 
133 #define OFFSET(x) offsetof(EBUR128Context, x)
134 #define A AV_OPT_FLAG_AUDIO_PARAM
135 #define V AV_OPT_FLAG_VIDEO_PARAM
136 #define F AV_OPT_FLAG_FILTERING_PARAM
137 static const AVOption ebur128_options[] = {
138  { "video", "set video output", OFFSET(do_video), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, V|F },
139  { "size", "set video size", OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "640x480"}, 0, 0, V|F },
140  { "meter", "set scale meter (+9 to +18)", OFFSET(meter), AV_OPT_TYPE_INT, {.i64 = 9}, 9, 18, V|F },
141  { "framelog", "force frame logging level", OFFSET(loglevel), AV_OPT_TYPE_INT, {.i64 = -1}, INT_MIN, INT_MAX, A|V|F, "level" },
142  { "info", "information logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_INFO}, INT_MIN, INT_MAX, A|V|F, "level" },
143  { "verbose", "verbose logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_VERBOSE}, INT_MIN, INT_MAX, A|V|F, "level" },
144  { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, A|V|F },
145  { NULL }
146 };
147 
148 AVFILTER_DEFINE_CLASS(ebur128);
149 
150 static const uint8_t graph_colors[] = {
151  0xdd, 0x66, 0x66, // value above 0LU non reached
152  0x66, 0x66, 0xdd, // value below 0LU non reached
153  0x96, 0x33, 0x33, // value above 0LU reached
154  0x33, 0x33, 0x96, // value below 0LU reached
155  0xdd, 0x96, 0x96, // value above 0LU line non reached
156  0x96, 0x96, 0xdd, // value below 0LU line non reached
157  0xdd, 0x33, 0x33, // value above 0LU line reached
158  0x33, 0x33, 0xdd, // value below 0LU line reached
159 };
160 
161 static const uint8_t *get_graph_color(const EBUR128Context *ebur128, int v, int y)
162 {
163  const int below0 = y > ebur128->y_zero_lu;
164  const int reached = y >= v;
165  const int line = ebur128->y_line_ref[y] || y == ebur128->y_zero_lu;
166  const int colorid = 4*line + 2*reached + below0;
167  return graph_colors + 3*colorid;
168 }
169 
170 static inline int lu_to_y(const EBUR128Context *ebur128, double v)
171 {
172  v += 2 * ebur128->meter; // make it in range [0;...]
173  v = av_clipf(v, 0, ebur128->scale_range); // make sure it's in the graph scale
174  v = ebur128->scale_range - v; // invert value (y=0 is on top)
175  return v * ebur128->graph.h / ebur128->scale_range; // rescale from scale range to px height
176 }
177 
178 #define FONT8 0
179 #define FONT16 1
180 
181 static const uint8_t font_colors[] = {
182  0xdd, 0xdd, 0x00,
183  0x00, 0x96, 0x96,
184 };
185 
186 static void drawtext(AVFrame *pic, int x, int y, int ftid, const uint8_t *color, const char *fmt, ...)
187 {
188  int i;
189  char buf[128] = {0};
190  const uint8_t *font;
191  int font_height;
192  va_list vl;
193 
194  if (ftid == FONT16) font = avpriv_vga16_font, font_height = 16;
195  else if (ftid == FONT8) font = avpriv_cga_font, font_height = 8;
196  else return;
197 
198  va_start(vl, fmt);
199  vsnprintf(buf, sizeof(buf), fmt, vl);
200  va_end(vl);
201 
202  for (i = 0; buf[i]; i++) {
203  int char_y, mask;
204  uint8_t *p = pic->data[0] + y*pic->linesize[0] + (x + i*8)*3;
205 
206  for (char_y = 0; char_y < font_height; char_y++) {
207  for (mask = 0x80; mask; mask >>= 1) {
208  if (font[buf[i] * font_height + char_y] & mask)
209  memcpy(p, color, 3);
210  else
211  memcpy(p, "\x00\x00\x00", 3);
212  p += 3;
213  }
214  p += pic->linesize[0] - 8*3;
215  }
216  }
217 }
218 
219 static void drawline(AVFrame *pic, int x, int y, int len, int step)
220 {
221  int i;
222  uint8_t *p = pic->data[0] + y*pic->linesize[0] + x*3;
223 
224  for (i = 0; i < len; i++) {
225  memcpy(p, "\x00\xff\x00", 3);
226  p += step;
227  }
228 }
229 
230 static int config_video_output(AVFilterLink *outlink)
231 {
232  int i, x, y;
233  uint8_t *p;
234  AVFilterContext *ctx = outlink->src;
235  EBUR128Context *ebur128 = ctx->priv;
236  AVFrame *outpicref;
237 
238  /* check if there is enough space to represent everything decently */
239  if (ebur128->w < 640 || ebur128->h < 480) {
240  av_log(ctx, AV_LOG_ERROR, "Video size %dx%d is too small, "
241  "minimum size is 640x480\n", ebur128->w, ebur128->h);
242  return AVERROR(EINVAL);
243  }
244  outlink->w = ebur128->w;
245  outlink->h = ebur128->h;
246 
247 #define PAD 8
248 
249  /* configure text area position and size */
250  ebur128->text.x = PAD;
251  ebur128->text.y = 40;
252  ebur128->text.w = 3 * 8; // 3 characters
253  ebur128->text.h = ebur128->h - PAD - ebur128->text.y;
254 
255  /* configure gauge position and size */
256  ebur128->gauge.w = 20;
257  ebur128->gauge.h = ebur128->text.h;
258  ebur128->gauge.x = ebur128->w - PAD - ebur128->gauge.w;
259  ebur128->gauge.y = ebur128->text.y;
260 
261  /* configure graph position and size */
262  ebur128->graph.x = ebur128->text.x + ebur128->text.w + PAD;
263  ebur128->graph.y = ebur128->gauge.y;
264  ebur128->graph.w = ebur128->gauge.x - ebur128->graph.x - PAD;
265  ebur128->graph.h = ebur128->gauge.h;
266 
267  /* graph and gauge share the LU-to-pixel code */
268  av_assert0(ebur128->graph.h == ebur128->gauge.h);
269 
270  /* prepare the initial picref buffer */
271  av_frame_free(&ebur128->outpicref);
272  ebur128->outpicref = outpicref =
273  ff_get_video_buffer(outlink, outlink->w, outlink->h);
274  if (!outpicref)
275  return AVERROR(ENOMEM);
276  outlink->sample_aspect_ratio = (AVRational){1,1};
277 
278  /* init y references values (to draw LU lines) */
279  ebur128->y_line_ref = av_calloc(ebur128->graph.h + 1, sizeof(*ebur128->y_line_ref));
280  if (!ebur128->y_line_ref)
281  return AVERROR(ENOMEM);
282 
283  /* black background */
284  memset(outpicref->data[0], 0, ebur128->h * outpicref->linesize[0]);
285 
286  /* draw LU legends */
287  drawtext(outpicref, PAD, PAD+16, FONT8, font_colors+3, " LU");
288  for (i = ebur128->meter; i >= -ebur128->meter * 2; i--) {
289  y = lu_to_y(ebur128, i);
290  x = PAD + (i < 10 && i > -10) * 8;
291  ebur128->y_line_ref[y] = i;
292  y -= 4; // -4 to center vertically
293  drawtext(outpicref, x, y + ebur128->graph.y, FONT8, font_colors+3,
294  "%c%d", i < 0 ? '-' : i > 0 ? '+' : ' ', FFABS(i));
295  }
296 
297  /* draw graph */
298  ebur128->y_zero_lu = lu_to_y(ebur128, 0);
299  p = outpicref->data[0] + ebur128->graph.y * outpicref->linesize[0]
300  + ebur128->graph.x * 3;
301  for (y = 0; y < ebur128->graph.h; y++) {
302  const uint8_t *c = get_graph_color(ebur128, INT_MAX, y);
303 
304  for (x = 0; x < ebur128->graph.w; x++)
305  memcpy(p + x*3, c, 3);
306  p += outpicref->linesize[0];
307  }
308 
309  /* draw fancy rectangles around the graph and the gauge */
310 #define DRAW_RECT(r) do { \
311  drawline(outpicref, r.x, r.y - 1, r.w, 3); \
312  drawline(outpicref, r.x, r.y + r.h, r.w, 3); \
313  drawline(outpicref, r.x - 1, r.y, r.h, outpicref->linesize[0]); \
314  drawline(outpicref, r.x + r.w, r.y, r.h, outpicref->linesize[0]); \
315 } while (0)
316  DRAW_RECT(ebur128->graph);
317  DRAW_RECT(ebur128->gauge);
318 
319  outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
320 
321  return 0;
322 }
323 
324 static int config_audio_input(AVFilterLink *inlink)
325 {
326  AVFilterContext *ctx = inlink->dst;
327  EBUR128Context *ebur128 = ctx->priv;
328 
329  /* force 100ms framing in case of metadata injection: the frames must have
330  * a granularity of the window overlap to be accurately exploited */
331  if (ebur128->metadata)
332  inlink->min_samples =
333  inlink->max_samples =
334  inlink->partial_buf_size = inlink->sample_rate / 10;
335  return 0;
336 }
337 
338 static int config_audio_output(AVFilterLink *outlink)
339 {
340  int i;
341  int idx_bitposn = 0;
342  AVFilterContext *ctx = outlink->src;
343  EBUR128Context *ebur128 = ctx->priv;
345 
346 #define BACK_MASK (AV_CH_BACK_LEFT |AV_CH_BACK_CENTER |AV_CH_BACK_RIGHT| \
347  AV_CH_TOP_BACK_LEFT|AV_CH_TOP_BACK_CENTER|AV_CH_TOP_BACK_RIGHT| \
348  AV_CH_SIDE_LEFT |AV_CH_SIDE_RIGHT| \
349  AV_CH_SURROUND_DIRECT_LEFT |AV_CH_SURROUND_DIRECT_RIGHT)
350 
351  ebur128->nb_channels = nb_channels;
352  ebur128->ch_weighting = av_calloc(nb_channels, sizeof(*ebur128->ch_weighting));
353  if (!ebur128->ch_weighting)
354  return AVERROR(ENOMEM);
355 
356  for (i = 0; i < nb_channels; i++) {
357 
358  /* find the next bit that is set starting from the right */
359  while ((outlink->channel_layout & 1ULL<<idx_bitposn) == 0 && idx_bitposn < 63)
360  idx_bitposn++;
361 
362  /* channel weighting */
363  if ((1ULL<<idx_bitposn & AV_CH_LOW_FREQUENCY) ||
364  (1ULL<<idx_bitposn & AV_CH_LOW_FREQUENCY_2)) {
365  ebur128->ch_weighting[i] = 0;
366  } else if (1ULL<<idx_bitposn & BACK_MASK) {
367  ebur128->ch_weighting[i] = 1.41;
368  } else {
369  ebur128->ch_weighting[i] = 1.0;
370  }
371 
372  idx_bitposn++;
373 
374  if (!ebur128->ch_weighting[i])
375  continue;
376 
377  /* bins buffer for the two integration window (400ms and 3s) */
378  ebur128->i400.cache[i] = av_calloc(I400_BINS, sizeof(*ebur128->i400.cache[0]));
379  ebur128->i3000.cache[i] = av_calloc(I3000_BINS, sizeof(*ebur128->i3000.cache[0]));
380  if (!ebur128->i400.cache[i] || !ebur128->i3000.cache[i])
381  return AVERROR(ENOMEM);
382  }
383 
384  outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
385 
386  return 0;
387 }
388 
389 #define ENERGY(loudness) (pow(10, ((loudness) + 0.691) / 10.))
390 #define LOUDNESS(energy) (-0.691 + 10 * log10(energy))
391 
392 static struct hist_entry *get_histogram(void)
393 {
394  int i;
395  struct hist_entry *h = av_calloc(HIST_SIZE, sizeof(*h));
396 
397  if (!h)
398  return NULL;
399  for (i = 0; i < HIST_SIZE; i++) {
400  h[i].loudness = i / (double)HIST_GRAIN + ABS_THRES;
401  h[i].energy = ENERGY(h[i].loudness);
402  }
403  return h;
404 }
405 
406 static av_cold int init(AVFilterContext *ctx)
407 {
408  EBUR128Context *ebur128 = ctx->priv;
409  AVFilterPad pad;
410 
411  if (ebur128->loglevel != AV_LOG_INFO &&
412  ebur128->loglevel != AV_LOG_VERBOSE) {
413  if (ebur128->do_video || ebur128->metadata)
414  ebur128->loglevel = AV_LOG_VERBOSE;
415  else
416  ebur128->loglevel = AV_LOG_INFO;
417  }
418 
419  // if meter is +9 scale, scale range is from -18 LU to +9 LU (or 3*9)
420  // if meter is +18 scale, scale range is from -36 LU to +18 LU (or 3*18)
421  ebur128->scale_range = 3 * ebur128->meter;
422 
423  ebur128->i400.histogram = get_histogram();
424  ebur128->i3000.histogram = get_histogram();
425  if (!ebur128->i400.histogram || !ebur128->i3000.histogram)
426  return AVERROR(ENOMEM);
427 
428  ebur128->integrated_loudness = ABS_THRES;
429  ebur128->loudness_range = 0;
430 
431  /* insert output pads */
432  if (ebur128->do_video) {
433  pad = (AVFilterPad){
434  .name = av_strdup("out0"),
435  .type = AVMEDIA_TYPE_VIDEO,
436  .config_props = config_video_output,
437  };
438  if (!pad.name)
439  return AVERROR(ENOMEM);
440  ff_insert_outpad(ctx, 0, &pad);
441  }
442  pad = (AVFilterPad){
443  .name = av_asprintf("out%d", ebur128->do_video),
444  .type = AVMEDIA_TYPE_AUDIO,
445  .config_props = config_audio_output,
446  };
447  if (!pad.name)
448  return AVERROR(ENOMEM);
449  ff_insert_outpad(ctx, ebur128->do_video, &pad);
450 
451  /* summary */
452  av_log(ctx, AV_LOG_VERBOSE, "EBU +%d scale\n", ebur128->meter);
453 
454  return 0;
455 }
456 
457 #define HIST_POS(power) (int)(((power) - ABS_THRES) * HIST_GRAIN)
458 
459 /* loudness and power should be set such as loudness = -0.691 +
460  * 10*log10(power), we just avoid doing that calculus two times */
461 static int gate_update(struct integrator *integ, double power,
462  double loudness, int gate_thres)
463 {
464  int ipower;
465  double relative_threshold;
466  int gate_hist_pos;
467 
468  /* update powers histograms by incrementing current power count */
469  ipower = av_clip(HIST_POS(loudness), 0, HIST_SIZE - 1);
470  integ->histogram[ipower].count++;
471 
472  /* compute relative threshold and get its position in the histogram */
473  integ->sum_kept_powers += power;
474  integ->nb_kept_powers++;
475  relative_threshold = integ->sum_kept_powers / integ->nb_kept_powers;
476  if (!relative_threshold)
477  relative_threshold = 1e-12;
478  integ->rel_threshold = LOUDNESS(relative_threshold) + gate_thres;
479  gate_hist_pos = av_clip(HIST_POS(integ->rel_threshold), 0, HIST_SIZE - 1);
480 
481  return gate_hist_pos;
482 }
483 
484 static int filter_frame(AVFilterLink *inlink, AVFrame *insamples)
485 {
486  int i, ch, idx_insample;
487  AVFilterContext *ctx = inlink->dst;
488  EBUR128Context *ebur128 = ctx->priv;
489  const int nb_channels = ebur128->nb_channels;
490  const int nb_samples = insamples->nb_samples;
491  const double *samples = (double *)insamples->data[0];
492  AVFrame *pic = ebur128->outpicref;
493 
494  for (idx_insample = 0; idx_insample < nb_samples; idx_insample++) {
495  const int bin_id_400 = ebur128->i400.cache_pos;
496  const int bin_id_3000 = ebur128->i3000.cache_pos;
497 
498 #define MOVE_TO_NEXT_CACHED_ENTRY(time) do { \
499  ebur128->i##time.cache_pos++; \
500  if (ebur128->i##time.cache_pos == I##time##_BINS) { \
501  ebur128->i##time.filled = 1; \
502  ebur128->i##time.cache_pos = 0; \
503  } \
504 } while (0)
505 
508 
509  for (ch = 0; ch < nb_channels; ch++) {
510  double bin;
511 
512  ebur128->x[ch * 3] = *samples++; // set X[i]
513 
514  if (!ebur128->ch_weighting[ch])
515  continue;
516 
517  /* Y[i] = X[i]*b0 + X[i-1]*b1 + X[i-2]*b2 - Y[i-1]*a1 - Y[i-2]*a2 */
518 #define FILTER(Y, X, name) do { \
519  double *dst = ebur128->Y + ch*3; \
520  double *src = ebur128->X + ch*3; \
521  dst[2] = dst[1]; \
522  dst[1] = dst[0]; \
523  dst[0] = src[0]*name##_B0 + src[1]*name##_B1 + src[2]*name##_B2 \
524  - dst[1]*name##_A1 - dst[2]*name##_A2; \
525 } while (0)
526 
527  // TODO: merge both filters in one?
528  FILTER(y, x, PRE); // apply pre-filter
529  ebur128->x[ch * 3 + 2] = ebur128->x[ch * 3 + 1];
530  ebur128->x[ch * 3 + 1] = ebur128->x[ch * 3 ];
531  FILTER(z, y, RLB); // apply RLB-filter
532 
533  bin = ebur128->z[ch * 3] * ebur128->z[ch * 3];
534 
535  /* add the new value, and limit the sum to the cache size (400ms or 3s)
536  * by removing the oldest one */
537  ebur128->i400.sum [ch] = ebur128->i400.sum [ch] + bin - ebur128->i400.cache [ch][bin_id_400];
538  ebur128->i3000.sum[ch] = ebur128->i3000.sum[ch] + bin - ebur128->i3000.cache[ch][bin_id_3000];
539 
540  /* override old cache entry with the new value */
541  ebur128->i400.cache [ch][bin_id_400 ] = bin;
542  ebur128->i3000.cache[ch][bin_id_3000] = bin;
543  }
544 
545  /* For integrated loudness, gating blocks are 400ms long with 75%
546  * overlap (see BS.1770-2 p5), so a re-computation is needed each 100ms
547  * (4800 samples at 48kHz). */
548  if (++ebur128->sample_count == 4800) {
549  double loudness_400, loudness_3000;
550  double power_400 = 1e-12, power_3000 = 1e-12;
551  AVFilterLink *outlink = ctx->outputs[0];
552  const int64_t pts = insamples->pts +
553  av_rescale_q(idx_insample, (AVRational){ 1, inlink->sample_rate },
554  outlink->time_base);
555 
556  ebur128->sample_count = 0;
557 
558 #define COMPUTE_LOUDNESS(m, time) do { \
559  if (ebur128->i##time.filled) { \
560  /* weighting sum of the last <time> ms */ \
561  for (ch = 0; ch < nb_channels; ch++) \
562  power_##time += ebur128->ch_weighting[ch] * ebur128->i##time.sum[ch]; \
563  power_##time /= I##time##_BINS; \
564  } \
565  loudness_##time = LOUDNESS(power_##time); \
566 } while (0)
567 
568  COMPUTE_LOUDNESS(M, 400);
569  COMPUTE_LOUDNESS(S, 3000);
570 
571  /* Integrated loudness */
572 #define I_GATE_THRES -10 // initially defined to -8 LU in the first EBU standard
573 
574  if (loudness_400 >= ABS_THRES) {
575  double integrated_sum = 0;
576  int nb_integrated = 0;
577  int gate_hist_pos = gate_update(&ebur128->i400, power_400,
578  loudness_400, I_GATE_THRES);
579 
580  /* compute integrated loudness by summing the histogram values
581  * above the relative threshold */
582  for (i = gate_hist_pos; i < HIST_SIZE; i++) {
583  const int nb_v = ebur128->i400.histogram[i].count;
584  nb_integrated += nb_v;
585  integrated_sum += nb_v * ebur128->i400.histogram[i].energy;
586  }
587  if (nb_integrated)
588  ebur128->integrated_loudness = LOUDNESS(integrated_sum / nb_integrated);
589  }
590 
591  /* LRA */
592 #define LRA_GATE_THRES -20
593 #define LRA_LOWER_PRC 10
594 #define LRA_HIGHER_PRC 95
595 
596  /* XXX: example code in EBU 3342 is ">=" but formula in BS.1770
597  * specs is ">" */
598  if (loudness_3000 >= ABS_THRES) {
599  int nb_powers = 0;
600  int gate_hist_pos = gate_update(&ebur128->i3000, power_3000,
601  loudness_3000, LRA_GATE_THRES);
602 
603  for (i = gate_hist_pos; i < HIST_SIZE; i++)
604  nb_powers += ebur128->i3000.histogram[i].count;
605  if (nb_powers) {
606  int n, nb_pow;
607 
608  /* get lower loudness to consider */
609  n = 0;
610  nb_pow = LRA_LOWER_PRC * nb_powers / 100. + 0.5;
611  for (i = gate_hist_pos; i < HIST_SIZE; i++) {
612  n += ebur128->i3000.histogram[i].count;
613  if (n >= nb_pow) {
614  ebur128->lra_low = ebur128->i3000.histogram[i].loudness;
615  break;
616  }
617  }
618 
619  /* get higher loudness to consider */
620  n = nb_powers;
621  nb_pow = LRA_HIGHER_PRC * nb_powers / 100. + 0.5;
622  for (i = HIST_SIZE - 1; i >= 0; i--) {
623  n -= ebur128->i3000.histogram[i].count;
624  if (n < nb_pow) {
625  ebur128->lra_high = ebur128->i3000.histogram[i].loudness;
626  break;
627  }
628  }
629 
630  // XXX: show low & high on the graph?
631  ebur128->loudness_range = ebur128->lra_high - ebur128->lra_low;
632  }
633  }
634 
635 #define LOG_FMT "M:%6.1f S:%6.1f I:%6.1f LUFS LRA:%6.1f LU"
636 
637  /* push one video frame */
638  if (ebur128->do_video) {
639  int x, y, ret;
640  uint8_t *p;
641 
642  const int y_loudness_lu_graph = lu_to_y(ebur128, loudness_3000 + 23);
643  const int y_loudness_lu_gauge = lu_to_y(ebur128, loudness_400 + 23);
644 
645  /* draw the graph using the short-term loudness */
646  p = pic->data[0] + ebur128->graph.y*pic->linesize[0] + ebur128->graph.x*3;
647  for (y = 0; y < ebur128->graph.h; y++) {
648  const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_graph, y);
649 
650  memmove(p, p + 3, (ebur128->graph.w - 1) * 3);
651  memcpy(p + (ebur128->graph.w - 1) * 3, c, 3);
652  p += pic->linesize[0];
653  }
654 
655  /* draw the gauge using the momentary loudness */
656  p = pic->data[0] + ebur128->gauge.y*pic->linesize[0] + ebur128->gauge.x*3;
657  for (y = 0; y < ebur128->gauge.h; y++) {
658  const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_gauge, y);
659 
660  for (x = 0; x < ebur128->gauge.w; x++)
661  memcpy(p + x*3, c, 3);
662  p += pic->linesize[0];
663  }
664 
665  /* draw textual info */
666  drawtext(pic, PAD, PAD - PAD/2, FONT16, font_colors,
667  LOG_FMT " ", // padding to erase trailing characters
668  loudness_400, loudness_3000,
669  ebur128->integrated_loudness, ebur128->loudness_range);
670 
671  /* set pts and push frame */
672  pic->pts = pts;
673  ret = ff_filter_frame(outlink, av_frame_clone(pic));
674  if (ret < 0)
675  return ret;
676  }
677 
678  if (ebur128->metadata) { /* happens only once per filter_frame call */
679  char metabuf[128];
680 #define SET_META(name, var) do { \
681  snprintf(metabuf, sizeof(metabuf), "%.3f", var); \
682  av_dict_set(&insamples->metadata, "lavfi.r128." name, metabuf, 0); \
683 } while (0)
684  SET_META("M", loudness_400);
685  SET_META("S", loudness_3000);
686  SET_META("I", ebur128->integrated_loudness);
687  SET_META("LRA", ebur128->loudness_range);
688  SET_META("LRA.low", ebur128->lra_low);
689  SET_META("LRA.high", ebur128->lra_high);
690  }
691 
692  av_log(ctx, ebur128->loglevel, "t: %-10s " LOG_FMT "\n",
693  av_ts2timestr(pts, &outlink->time_base),
694  loudness_400, loudness_3000,
695  ebur128->integrated_loudness, ebur128->loudness_range);
696  }
697  }
698 
699  return ff_filter_frame(ctx->outputs[ebur128->do_video], insamples);
700 }
701 
703 {
704  EBUR128Context *ebur128 = ctx->priv;
707  AVFilterLink *inlink = ctx->inputs[0];
708  AVFilterLink *outlink = ctx->outputs[0];
709 
711  static const int input_srate[] = {48000, -1}; // ITU-R BS.1770 provides coeff only for 48kHz
712  static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_RGB24, AV_PIX_FMT_NONE };
713 
714  /* set optional output video format */
715  if (ebur128->do_video) {
716  formats = ff_make_format_list(pix_fmts);
717  if (!formats)
718  return AVERROR(ENOMEM);
719  ff_formats_ref(formats, &outlink->in_formats);
720  outlink = ctx->outputs[1];
721  }
722 
723  /* set input and output audio formats
724  * Note: ff_set_common_* functions are not used because they affect all the
725  * links, and thus break the video format negotiation */
726  formats = ff_make_format_list(sample_fmts);
727  if (!formats)
728  return AVERROR(ENOMEM);
729  ff_formats_ref(formats, &inlink->out_formats);
730  ff_formats_ref(formats, &outlink->in_formats);
731 
732  layouts = ff_all_channel_layouts();
733  if (!layouts)
734  return AVERROR(ENOMEM);
735  ff_channel_layouts_ref(layouts, &inlink->out_channel_layouts);
736  ff_channel_layouts_ref(layouts, &outlink->in_channel_layouts);
737 
738  formats = ff_make_format_list(input_srate);
739  if (!formats)
740  return AVERROR(ENOMEM);
741  ff_formats_ref(formats, &inlink->out_samplerates);
742  ff_formats_ref(formats, &outlink->in_samplerates);
743 
744  return 0;
745 }
746 
747 static av_cold void uninit(AVFilterContext *ctx)
748 {
749  int i;
750  EBUR128Context *ebur128 = ctx->priv;
751 
752  av_log(ctx, AV_LOG_INFO, "Summary:\n\n"
753  " Integrated loudness:\n"
754  " I: %5.1f LUFS\n"
755  " Threshold: %5.1f LUFS\n\n"
756  " Loudness range:\n"
757  " LRA: %5.1f LU\n"
758  " Threshold: %5.1f LUFS\n"
759  " LRA low: %5.1f LUFS\n"
760  " LRA high: %5.1f LUFS\n",
761  ebur128->integrated_loudness, ebur128->i400.rel_threshold,
762  ebur128->loudness_range, ebur128->i3000.rel_threshold,
763  ebur128->lra_low, ebur128->lra_high);
764 
765  av_freep(&ebur128->y_line_ref);
766  av_freep(&ebur128->ch_weighting);
767  av_freep(&ebur128->i400.histogram);
768  av_freep(&ebur128->i3000.histogram);
769  for (i = 0; i < ebur128->nb_channels; i++) {
770  av_freep(&ebur128->i400.cache[i]);
771  av_freep(&ebur128->i3000.cache[i]);
772  }
773  for (i = 0; i < ctx->nb_outputs; i++)
774  av_freep(&ctx->output_pads[i].name);
775  av_frame_free(&ebur128->outpicref);
776 }
777 
778 static const AVFilterPad ebur128_inputs[] = {
779  {
780  .name = "default",
781  .type = AVMEDIA_TYPE_AUDIO,
782  .filter_frame = filter_frame,
783  .config_props = config_audio_input,
784  },
785  { NULL }
786 };
787 
789  .name = "ebur128",
790  .description = NULL_IF_CONFIG_SMALL("EBU R128 scanner."),
791  .priv_size = sizeof(EBUR128Context),
792  .init = init,
793  .uninit = uninit,
795  .inputs = ebur128_inputs,
796  .outputs = NULL,
797  .priv_class = &ebur128_class,
799 };