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 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"
35 #include "libavutil/dict.h"
37 #include "libavutil/opt.h"
38 #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  /* 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  double *swr_buf; ///< resampled audio data for true peak metering
103  int swr_linesize;
104 #endif
105 
106  /* video */
107  int do_video; ///< 1 if video output enabled, 0 otherwise
108  int w, h; ///< size of the video output
109  struct rect text; ///< rectangle for the LU legend on the left
110  struct rect graph; ///< rectangle for the main graph in the center
111  struct rect gauge; ///< rectangle for the gauge on the right
112  AVFrame *outpicref; ///< output picture reference, updated regularly
113  int meter; ///< select a EBU mode between +9 and +18
114  int scale_range; ///< the range of LU values according to the meter
115  int y_zero_lu; ///< the y value (pixel position) for 0 LU
116  int *y_line_ref; ///< y reference values for drawing the LU lines in the graph and the gauge
117 
118  /* audio */
119  int nb_channels; ///< number of channels in the input
120  double *ch_weighting; ///< channel weighting mapping
121  int sample_count; ///< sample count used for refresh frequency, reset at refresh
122 
123  /* Filter caches.
124  * The mult by 3 in the following is for X[i], X[i-1] and X[i-2] */
125  double x[MAX_CHANNELS * 3]; ///< 3 input samples cache for each channel
126  double y[MAX_CHANNELS * 3]; ///< 3 pre-filter samples cache for each channel
127  double z[MAX_CHANNELS * 3]; ///< 3 RLB-filter samples cache for each channel
128 
129 #define I400_BINS (48000 * 4 / 10)
130 #define I3000_BINS (48000 * 3)
131  struct integrator i400; ///< 400ms integrator, used for Momentary loudness (M), and Integrated loudness (I)
132  struct integrator i3000; ///< 3s integrator, used for Short term loudness (S), and Loudness Range (LRA)
133 
134  /* I and LRA specific */
135  double integrated_loudness; ///< integrated loudness in LUFS (I)
136  double loudness_range; ///< loudness range in LU (LRA)
137  double lra_low, lra_high; ///< low and high LRA values
138 
139  /* misc */
140  int loglevel; ///< log level for frame logging
141  int metadata; ///< whether or not to inject loudness results in frames
143 
144 enum {
148 };
149 
150 #define OFFSET(x) offsetof(EBUR128Context, x)
151 #define A AV_OPT_FLAG_AUDIO_PARAM
152 #define V AV_OPT_FLAG_VIDEO_PARAM
153 #define F AV_OPT_FLAG_FILTERING_PARAM
154 static const AVOption ebur128_options[] = {
155  { "video", "set video output", OFFSET(do_video), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, V|F },
156  { "size", "set video size", OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "640x480"}, 0, 0, V|F },
157  { "meter", "set scale meter (+9 to +18)", OFFSET(meter), AV_OPT_TYPE_INT, {.i64 = 9}, 9, 18, V|F },
158  { "framelog", "force frame logging level", OFFSET(loglevel), AV_OPT_TYPE_INT, {.i64 = -1}, INT_MIN, INT_MAX, A|V|F, "level" },
159  { "info", "information logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_INFO}, INT_MIN, INT_MAX, A|V|F, "level" },
160  { "verbose", "verbose logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_VERBOSE}, INT_MIN, INT_MAX, A|V|F, "level" },
161  { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, A|V|F },
162  { "peak", "set peak mode", OFFSET(peak_mode), AV_OPT_TYPE_FLAGS, {.i64 = PEAK_MODE_NONE}, 0, INT_MAX, A|F, "mode" },
163  { "none", "disable any peak mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_NONE}, INT_MIN, INT_MAX, A|F, "mode" },
164  { "sample", "enable peak-sample mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_SAMPLES_PEAKS}, INT_MIN, INT_MAX, A|F, "mode" },
165  { "true", "enable true-peak mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_TRUE_PEAKS}, INT_MIN, INT_MAX, A|F, "mode" },
166  { NULL },
167 };
168 
169 AVFILTER_DEFINE_CLASS(ebur128);
170 
171 static const uint8_t graph_colors[] = {
172  0xdd, 0x66, 0x66, // value above 0LU non reached
173  0x66, 0x66, 0xdd, // value below 0LU non reached
174  0x96, 0x33, 0x33, // value above 0LU reached
175  0x33, 0x33, 0x96, // value below 0LU reached
176  0xdd, 0x96, 0x96, // value above 0LU line non reached
177  0x96, 0x96, 0xdd, // value below 0LU line non reached
178  0xdd, 0x33, 0x33, // value above 0LU line reached
179  0x33, 0x33, 0xdd, // value below 0LU line reached
180 };
181 
182 static const uint8_t *get_graph_color(const EBUR128Context *ebur128, int v, int y)
183 {
184  const int below0 = y > ebur128->y_zero_lu;
185  const int reached = y >= v;
186  const int line = ebur128->y_line_ref[y] || y == ebur128->y_zero_lu;
187  const int colorid = 4*line + 2*reached + below0;
188  return graph_colors + 3*colorid;
189 }
190 
191 static inline int lu_to_y(const EBUR128Context *ebur128, double v)
192 {
193  v += 2 * ebur128->meter; // make it in range [0;...]
194  v = av_clipf(v, 0, ebur128->scale_range); // make sure it's in the graph scale
195  v = ebur128->scale_range - v; // invert value (y=0 is on top)
196  return v * ebur128->graph.h / ebur128->scale_range; // rescale from scale range to px height
197 }
198 
199 #define FONT8 0
200 #define FONT16 1
201 
202 static const uint8_t font_colors[] = {
203  0xdd, 0xdd, 0x00,
204  0x00, 0x96, 0x96,
205 };
206 
207 static void drawtext(AVFrame *pic, int x, int y, int ftid, const uint8_t *color, const char *fmt, ...)
208 {
209  int i;
210  char buf[128] = {0};
211  const uint8_t *font;
212  int font_height;
213  va_list vl;
214 
215  if (ftid == FONT16) font = avpriv_vga16_font, font_height = 16;
216  else if (ftid == FONT8) font = avpriv_cga_font, font_height = 8;
217  else return;
218 
219  va_start(vl, fmt);
220  vsnprintf(buf, sizeof(buf), fmt, vl);
221  va_end(vl);
222 
223  for (i = 0; buf[i]; i++) {
224  int char_y, mask;
225  uint8_t *p = pic->data[0] + y*pic->linesize[0] + (x + i*8)*3;
226 
227  for (char_y = 0; char_y < font_height; char_y++) {
228  for (mask = 0x80; mask; mask >>= 1) {
229  if (font[buf[i] * font_height + char_y] & mask)
230  memcpy(p, color, 3);
231  else
232  memcpy(p, "\x00\x00\x00", 3);
233  p += 3;
234  }
235  p += pic->linesize[0] - 8*3;
236  }
237  }
238 }
239 
240 static void drawline(AVFrame *pic, int x, int y, int len, int step)
241 {
242  int i;
243  uint8_t *p = pic->data[0] + y*pic->linesize[0] + x*3;
244 
245  for (i = 0; i < len; i++) {
246  memcpy(p, "\x00\xff\x00", 3);
247  p += step;
248  }
249 }
250 
251 static int config_video_output(AVFilterLink *outlink)
252 {
253  int i, x, y;
254  uint8_t *p;
255  AVFilterContext *ctx = outlink->src;
256  EBUR128Context *ebur128 = ctx->priv;
257  AVFrame *outpicref;
258 
259  /* check if there is enough space to represent everything decently */
260  if (ebur128->w < 640 || ebur128->h < 480) {
261  av_log(ctx, AV_LOG_ERROR, "Video size %dx%d is too small, "
262  "minimum size is 640x480\n", ebur128->w, ebur128->h);
263  return AVERROR(EINVAL);
264  }
265  outlink->w = ebur128->w;
266  outlink->h = ebur128->h;
267 
268 #define PAD 8
269 
270  /* configure text area position and size */
271  ebur128->text.x = PAD;
272  ebur128->text.y = 40;
273  ebur128->text.w = 3 * 8; // 3 characters
274  ebur128->text.h = ebur128->h - PAD - ebur128->text.y;
275 
276  /* configure gauge position and size */
277  ebur128->gauge.w = 20;
278  ebur128->gauge.h = ebur128->text.h;
279  ebur128->gauge.x = ebur128->w - PAD - ebur128->gauge.w;
280  ebur128->gauge.y = ebur128->text.y;
281 
282  /* configure graph position and size */
283  ebur128->graph.x = ebur128->text.x + ebur128->text.w + PAD;
284  ebur128->graph.y = ebur128->gauge.y;
285  ebur128->graph.w = ebur128->gauge.x - ebur128->graph.x - PAD;
286  ebur128->graph.h = ebur128->gauge.h;
287 
288  /* graph and gauge share the LU-to-pixel code */
289  av_assert0(ebur128->graph.h == ebur128->gauge.h);
290 
291  /* prepare the initial picref buffer */
292  av_frame_free(&ebur128->outpicref);
293  ebur128->outpicref = outpicref =
294  ff_get_video_buffer(outlink, outlink->w, outlink->h);
295  if (!outpicref)
296  return AVERROR(ENOMEM);
297  outlink->sample_aspect_ratio = (AVRational){1,1};
298 
299  /* init y references values (to draw LU lines) */
300  ebur128->y_line_ref = av_calloc(ebur128->graph.h + 1, sizeof(*ebur128->y_line_ref));
301  if (!ebur128->y_line_ref)
302  return AVERROR(ENOMEM);
303 
304  /* black background */
305  memset(outpicref->data[0], 0, ebur128->h * outpicref->linesize[0]);
306 
307  /* draw LU legends */
308  drawtext(outpicref, PAD, PAD+16, FONT8, font_colors+3, " LU");
309  for (i = ebur128->meter; i >= -ebur128->meter * 2; i--) {
310  y = lu_to_y(ebur128, i);
311  x = PAD + (i < 10 && i > -10) * 8;
312  ebur128->y_line_ref[y] = i;
313  y -= 4; // -4 to center vertically
314  drawtext(outpicref, x, y + ebur128->graph.y, FONT8, font_colors+3,
315  "%c%d", i < 0 ? '-' : i > 0 ? '+' : ' ', FFABS(i));
316  }
317 
318  /* draw graph */
319  ebur128->y_zero_lu = lu_to_y(ebur128, 0);
320  p = outpicref->data[0] + ebur128->graph.y * outpicref->linesize[0]
321  + ebur128->graph.x * 3;
322  for (y = 0; y < ebur128->graph.h; y++) {
323  const uint8_t *c = get_graph_color(ebur128, INT_MAX, y);
324 
325  for (x = 0; x < ebur128->graph.w; x++)
326  memcpy(p + x*3, c, 3);
327  p += outpicref->linesize[0];
328  }
329 
330  /* draw fancy rectangles around the graph and the gauge */
331 #define DRAW_RECT(r) do { \
332  drawline(outpicref, r.x, r.y - 1, r.w, 3); \
333  drawline(outpicref, r.x, r.y + r.h, r.w, 3); \
334  drawline(outpicref, r.x - 1, r.y, r.h, outpicref->linesize[0]); \
335  drawline(outpicref, r.x + r.w, r.y, r.h, outpicref->linesize[0]); \
336 } while (0)
337  DRAW_RECT(ebur128->graph);
338  DRAW_RECT(ebur128->gauge);
339 
340  outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
341 
342  return 0;
343 }
344 
345 static int config_audio_input(AVFilterLink *inlink)
346 {
347  AVFilterContext *ctx = inlink->dst;
348  EBUR128Context *ebur128 = ctx->priv;
349 
350  /* Force 100ms framing in case of metadata injection: the frames must have
351  * a granularity of the window overlap to be accurately exploited.
352  * As for the true peaks mode, it just simplifies the resampling buffer
353  * allocation and the lookup in it (since sample buffers differ in size, it
354  * can be more complex to integrate in the one-sample loop of
355  * filter_frame()). */
356  if (ebur128->metadata || (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS))
357  inlink->min_samples =
358  inlink->max_samples =
359  inlink->partial_buf_size = inlink->sample_rate / 10;
360  return 0;
361 }
362 
363 static int config_audio_output(AVFilterLink *outlink)
364 {
365  int i;
366  int idx_bitposn = 0;
367  AVFilterContext *ctx = outlink->src;
368  EBUR128Context *ebur128 = ctx->priv;
370 
371 #define BACK_MASK (AV_CH_BACK_LEFT |AV_CH_BACK_CENTER |AV_CH_BACK_RIGHT| \
372  AV_CH_TOP_BACK_LEFT|AV_CH_TOP_BACK_CENTER|AV_CH_TOP_BACK_RIGHT| \
373  AV_CH_SIDE_LEFT |AV_CH_SIDE_RIGHT| \
374  AV_CH_SURROUND_DIRECT_LEFT |AV_CH_SURROUND_DIRECT_RIGHT)
375 
376  ebur128->nb_channels = nb_channels;
377  ebur128->ch_weighting = av_calloc(nb_channels, sizeof(*ebur128->ch_weighting));
378  if (!ebur128->ch_weighting)
379  return AVERROR(ENOMEM);
380 
381  for (i = 0; i < nb_channels; i++) {
382 
383  /* find the next bit that is set starting from the right */
384  while ((outlink->channel_layout & 1ULL<<idx_bitposn) == 0 && idx_bitposn < 63)
385  idx_bitposn++;
386 
387  /* channel weighting */
388  if ((1ULL<<idx_bitposn & AV_CH_LOW_FREQUENCY) ||
389  (1ULL<<idx_bitposn & AV_CH_LOW_FREQUENCY_2)) {
390  ebur128->ch_weighting[i] = 0;
391  } else if (1ULL<<idx_bitposn & BACK_MASK) {
392  ebur128->ch_weighting[i] = 1.41;
393  } else {
394  ebur128->ch_weighting[i] = 1.0;
395  }
396 
397  idx_bitposn++;
398 
399  if (!ebur128->ch_weighting[i])
400  continue;
401 
402  /* bins buffer for the two integration window (400ms and 3s) */
403  ebur128->i400.cache[i] = av_calloc(I400_BINS, sizeof(*ebur128->i400.cache[0]));
404  ebur128->i3000.cache[i] = av_calloc(I3000_BINS, sizeof(*ebur128->i3000.cache[0]));
405  if (!ebur128->i400.cache[i] || !ebur128->i3000.cache[i])
406  return AVERROR(ENOMEM);
407  }
408 
409  outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
410 
411 #if CONFIG_SWRESAMPLE
412  if (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS) {
413  int ret;
414 
415  ebur128->swr_buf = av_malloc(19200 * nb_channels * sizeof(double));
416  ebur128->true_peaks = av_calloc(nb_channels, sizeof(*ebur128->true_peaks));
417  ebur128->true_peaks_per_frame = av_calloc(nb_channels, sizeof(*ebur128->true_peaks_per_frame));
418  ebur128->swr_ctx = swr_alloc();
419  if (!ebur128->swr_buf || !ebur128->true_peaks ||
420  !ebur128->true_peaks_per_frame || !ebur128->swr_ctx)
421  return AVERROR(ENOMEM);
422 
423  av_opt_set_int(ebur128->swr_ctx, "in_channel_layout", outlink->channel_layout, 0);
424  av_opt_set_int(ebur128->swr_ctx, "in_sample_rate", outlink->sample_rate, 0);
425  av_opt_set_sample_fmt(ebur128->swr_ctx, "in_sample_fmt", outlink->format, 0);
426 
427  av_opt_set_int(ebur128->swr_ctx, "out_channel_layout", outlink->channel_layout, 0);
428  av_opt_set_int(ebur128->swr_ctx, "out_sample_rate", 192000, 0);
429  av_opt_set_sample_fmt(ebur128->swr_ctx, "out_sample_fmt", outlink->format, 0);
430 
431  ret = swr_init(ebur128->swr_ctx);
432  if (ret < 0)
433  return ret;
434  }
435 #endif
436 
437  if (ebur128->peak_mode & PEAK_MODE_SAMPLES_PEAKS) {
438  ebur128->sample_peaks = av_calloc(nb_channels, sizeof(*ebur128->sample_peaks));
439  if (!ebur128->sample_peaks)
440  return AVERROR(ENOMEM);
441  }
442 
443  return 0;
444 }
445 
446 #define ENERGY(loudness) (pow(10, ((loudness) + 0.691) / 10.))
447 #define LOUDNESS(energy) (-0.691 + 10 * log10(energy))
448 #define DBFS(energy) (20 * log10(energy))
449 
450 static struct hist_entry *get_histogram(void)
451 {
452  int i;
453  struct hist_entry *h = av_calloc(HIST_SIZE, sizeof(*h));
454 
455  if (!h)
456  return NULL;
457  for (i = 0; i < HIST_SIZE; i++) {
458  h[i].loudness = i / (double)HIST_GRAIN + ABS_THRES;
459  h[i].energy = ENERGY(h[i].loudness);
460  }
461  return h;
462 }
463 
464 static av_cold int init(AVFilterContext *ctx)
465 {
466  EBUR128Context *ebur128 = ctx->priv;
467  AVFilterPad pad;
468 
469  if (ebur128->loglevel != AV_LOG_INFO &&
470  ebur128->loglevel != AV_LOG_VERBOSE) {
471  if (ebur128->do_video || ebur128->metadata)
472  ebur128->loglevel = AV_LOG_VERBOSE;
473  else
474  ebur128->loglevel = AV_LOG_INFO;
475  }
476 
477  if (!CONFIG_SWRESAMPLE && (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS)) {
478  av_log(ctx, AV_LOG_ERROR,
479  "True-peak mode requires libswresample to be performed\n");
480  return AVERROR(EINVAL);
481  }
482 
483  // if meter is +9 scale, scale range is from -18 LU to +9 LU (or 3*9)
484  // if meter is +18 scale, scale range is from -36 LU to +18 LU (or 3*18)
485  ebur128->scale_range = 3 * ebur128->meter;
486 
487  ebur128->i400.histogram = get_histogram();
488  ebur128->i3000.histogram = get_histogram();
489  if (!ebur128->i400.histogram || !ebur128->i3000.histogram)
490  return AVERROR(ENOMEM);
491 
492  ebur128->integrated_loudness = ABS_THRES;
493  ebur128->loudness_range = 0;
494 
495  /* insert output pads */
496  if (ebur128->do_video) {
497  pad = (AVFilterPad){
498  .name = av_strdup("out0"),
499  .type = AVMEDIA_TYPE_VIDEO,
500  .config_props = config_video_output,
501  };
502  if (!pad.name)
503  return AVERROR(ENOMEM);
504  ff_insert_outpad(ctx, 0, &pad);
505  }
506  pad = (AVFilterPad){
507  .name = av_asprintf("out%d", ebur128->do_video),
508  .type = AVMEDIA_TYPE_AUDIO,
509  .config_props = config_audio_output,
510  };
511  if (!pad.name)
512  return AVERROR(ENOMEM);
513  ff_insert_outpad(ctx, ebur128->do_video, &pad);
514 
515  /* summary */
516  av_log(ctx, AV_LOG_VERBOSE, "EBU +%d scale\n", ebur128->meter);
517 
518  return 0;
519 }
520 
521 #define HIST_POS(power) (int)(((power) - ABS_THRES) * HIST_GRAIN)
522 
523 /* loudness and power should be set such as loudness = -0.691 +
524  * 10*log10(power), we just avoid doing that calculus two times */
525 static int gate_update(struct integrator *integ, double power,
526  double loudness, int gate_thres)
527 {
528  int ipower;
529  double relative_threshold;
530  int gate_hist_pos;
531 
532  /* update powers histograms by incrementing current power count */
533  ipower = av_clip(HIST_POS(loudness), 0, HIST_SIZE - 1);
534  integ->histogram[ipower].count++;
535 
536  /* compute relative threshold and get its position in the histogram */
537  integ->sum_kept_powers += power;
538  integ->nb_kept_powers++;
539  relative_threshold = integ->sum_kept_powers / integ->nb_kept_powers;
540  if (!relative_threshold)
541  relative_threshold = 1e-12;
542  integ->rel_threshold = LOUDNESS(relative_threshold) + gate_thres;
543  gate_hist_pos = av_clip(HIST_POS(integ->rel_threshold), 0, HIST_SIZE - 1);
544 
545  return gate_hist_pos;
546 }
547 
548 static int filter_frame(AVFilterLink *inlink, AVFrame *insamples)
549 {
550  int i, ch, idx_insample;
551  AVFilterContext *ctx = inlink->dst;
552  EBUR128Context *ebur128 = ctx->priv;
553  const int nb_channels = ebur128->nb_channels;
554  const int nb_samples = insamples->nb_samples;
555  const double *samples = (double *)insamples->data[0];
556  AVFrame *pic = ebur128->outpicref;
557 
558 #if CONFIG_SWRESAMPLE
559  if (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS) {
560  const double *swr_samples = ebur128->swr_buf;
561  int ret = swr_convert(ebur128->swr_ctx, (uint8_t**)&ebur128->swr_buf, 19200,
562  (const uint8_t **)insamples->data, nb_samples);
563  if (ret < 0)
564  return ret;
565  for (ch = 0; ch < nb_channels; ch++)
566  ebur128->true_peaks_per_frame[ch] = 0.0;
567  for (idx_insample = 0; idx_insample < ret; idx_insample++) {
568  for (ch = 0; ch < nb_channels; ch++) {
569  ebur128->true_peaks[ch] = FFMAX(ebur128->true_peaks[ch], FFABS(*swr_samples));
570  ebur128->true_peaks_per_frame[ch] = FFMAX(ebur128->true_peaks_per_frame[ch],
571  FFABS(*swr_samples));
572  swr_samples++;
573  }
574  }
575  }
576 #endif
577 
578  for (idx_insample = 0; idx_insample < nb_samples; idx_insample++) {
579  const int bin_id_400 = ebur128->i400.cache_pos;
580  const int bin_id_3000 = ebur128->i3000.cache_pos;
581 
582 #define MOVE_TO_NEXT_CACHED_ENTRY(time) do { \
583  ebur128->i##time.cache_pos++; \
584  if (ebur128->i##time.cache_pos == I##time##_BINS) { \
585  ebur128->i##time.filled = 1; \
586  ebur128->i##time.cache_pos = 0; \
587  } \
588 } while (0)
589 
592 
593  for (ch = 0; ch < nb_channels; ch++) {
594  double bin;
595 
596  if (ebur128->peak_mode & PEAK_MODE_SAMPLES_PEAKS)
597  ebur128->sample_peaks[ch] = FFMAX(ebur128->sample_peaks[ch], FFABS(*samples));
598 
599  ebur128->x[ch * 3] = *samples++; // set X[i]
600 
601  if (!ebur128->ch_weighting[ch])
602  continue;
603 
604  /* Y[i] = X[i]*b0 + X[i-1]*b1 + X[i-2]*b2 - Y[i-1]*a1 - Y[i-2]*a2 */
605 #define FILTER(Y, X, name) do { \
606  double *dst = ebur128->Y + ch*3; \
607  double *src = ebur128->X + ch*3; \
608  dst[2] = dst[1]; \
609  dst[1] = dst[0]; \
610  dst[0] = src[0]*name##_B0 + src[1]*name##_B1 + src[2]*name##_B2 \
611  - dst[1]*name##_A1 - dst[2]*name##_A2; \
612 } while (0)
613 
614  // TODO: merge both filters in one?
615  FILTER(y, x, PRE); // apply pre-filter
616  ebur128->x[ch * 3 + 2] = ebur128->x[ch * 3 + 1];
617  ebur128->x[ch * 3 + 1] = ebur128->x[ch * 3 ];
618  FILTER(z, y, RLB); // apply RLB-filter
619 
620  bin = ebur128->z[ch * 3] * ebur128->z[ch * 3];
621 
622  /* add the new value, and limit the sum to the cache size (400ms or 3s)
623  * by removing the oldest one */
624  ebur128->i400.sum [ch] = ebur128->i400.sum [ch] + bin - ebur128->i400.cache [ch][bin_id_400];
625  ebur128->i3000.sum[ch] = ebur128->i3000.sum[ch] + bin - ebur128->i3000.cache[ch][bin_id_3000];
626 
627  /* override old cache entry with the new value */
628  ebur128->i400.cache [ch][bin_id_400 ] = bin;
629  ebur128->i3000.cache[ch][bin_id_3000] = bin;
630  }
631 
632  /* For integrated loudness, gating blocks are 400ms long with 75%
633  * overlap (see BS.1770-2 p5), so a re-computation is needed each 100ms
634  * (4800 samples at 48kHz). */
635  if (++ebur128->sample_count == 4800) {
636  double loudness_400, loudness_3000;
637  double power_400 = 1e-12, power_3000 = 1e-12;
638  AVFilterLink *outlink = ctx->outputs[0];
639  const int64_t pts = insamples->pts +
640  av_rescale_q(idx_insample, (AVRational){ 1, inlink->sample_rate },
641  outlink->time_base);
642 
643  ebur128->sample_count = 0;
644 
645 #define COMPUTE_LOUDNESS(m, time) do { \
646  if (ebur128->i##time.filled) { \
647  /* weighting sum of the last <time> ms */ \
648  for (ch = 0; ch < nb_channels; ch++) \
649  power_##time += ebur128->ch_weighting[ch] * ebur128->i##time.sum[ch]; \
650  power_##time /= I##time##_BINS; \
651  } \
652  loudness_##time = LOUDNESS(power_##time); \
653 } while (0)
654 
655  COMPUTE_LOUDNESS(M, 400);
656  COMPUTE_LOUDNESS(S, 3000);
657 
658  /* Integrated loudness */
659 #define I_GATE_THRES -10 // initially defined to -8 LU in the first EBU standard
660 
661  if (loudness_400 >= ABS_THRES) {
662  double integrated_sum = 0;
663  int nb_integrated = 0;
664  int gate_hist_pos = gate_update(&ebur128->i400, power_400,
665  loudness_400, I_GATE_THRES);
666 
667  /* compute integrated loudness by summing the histogram values
668  * above the relative threshold */
669  for (i = gate_hist_pos; i < HIST_SIZE; i++) {
670  const int nb_v = ebur128->i400.histogram[i].count;
671  nb_integrated += nb_v;
672  integrated_sum += nb_v * ebur128->i400.histogram[i].energy;
673  }
674  if (nb_integrated)
675  ebur128->integrated_loudness = LOUDNESS(integrated_sum / nb_integrated);
676  }
677 
678  /* LRA */
679 #define LRA_GATE_THRES -20
680 #define LRA_LOWER_PRC 10
681 #define LRA_HIGHER_PRC 95
682 
683  /* XXX: example code in EBU 3342 is ">=" but formula in BS.1770
684  * specs is ">" */
685  if (loudness_3000 >= ABS_THRES) {
686  int nb_powers = 0;
687  int gate_hist_pos = gate_update(&ebur128->i3000, power_3000,
688  loudness_3000, LRA_GATE_THRES);
689 
690  for (i = gate_hist_pos; i < HIST_SIZE; i++)
691  nb_powers += ebur128->i3000.histogram[i].count;
692  if (nb_powers) {
693  int n, nb_pow;
694 
695  /* get lower loudness to consider */
696  n = 0;
697  nb_pow = LRA_LOWER_PRC * nb_powers / 100. + 0.5;
698  for (i = gate_hist_pos; i < HIST_SIZE; i++) {
699  n += ebur128->i3000.histogram[i].count;
700  if (n >= nb_pow) {
701  ebur128->lra_low = ebur128->i3000.histogram[i].loudness;
702  break;
703  }
704  }
705 
706  /* get higher loudness to consider */
707  n = nb_powers;
708  nb_pow = LRA_HIGHER_PRC * nb_powers / 100. + 0.5;
709  for (i = HIST_SIZE - 1; i >= 0; i--) {
710  n -= ebur128->i3000.histogram[i].count;
711  if (n < nb_pow) {
712  ebur128->lra_high = ebur128->i3000.histogram[i].loudness;
713  break;
714  }
715  }
716 
717  // XXX: show low & high on the graph?
718  ebur128->loudness_range = ebur128->lra_high - ebur128->lra_low;
719  }
720  }
721 
722 #define LOG_FMT "M:%6.1f S:%6.1f I:%6.1f LUFS LRA:%6.1f LU"
723 
724  /* push one video frame */
725  if (ebur128->do_video) {
726  int x, y, ret;
727  uint8_t *p;
728 
729  const int y_loudness_lu_graph = lu_to_y(ebur128, loudness_3000 + 23);
730  const int y_loudness_lu_gauge = lu_to_y(ebur128, loudness_400 + 23);
731 
732  /* draw the graph using the short-term loudness */
733  p = pic->data[0] + ebur128->graph.y*pic->linesize[0] + ebur128->graph.x*3;
734  for (y = 0; y < ebur128->graph.h; y++) {
735  const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_graph, y);
736 
737  memmove(p, p + 3, (ebur128->graph.w - 1) * 3);
738  memcpy(p + (ebur128->graph.w - 1) * 3, c, 3);
739  p += pic->linesize[0];
740  }
741 
742  /* draw the gauge using the momentary loudness */
743  p = pic->data[0] + ebur128->gauge.y*pic->linesize[0] + ebur128->gauge.x*3;
744  for (y = 0; y < ebur128->gauge.h; y++) {
745  const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_gauge, y);
746 
747  for (x = 0; x < ebur128->gauge.w; x++)
748  memcpy(p + x*3, c, 3);
749  p += pic->linesize[0];
750  }
751 
752  /* draw textual info */
753  drawtext(pic, PAD, PAD - PAD/2, FONT16, font_colors,
754  LOG_FMT " ", // padding to erase trailing characters
755  loudness_400, loudness_3000,
756  ebur128->integrated_loudness, ebur128->loudness_range);
757 
758  /* set pts and push frame */
759  pic->pts = pts;
760  ret = ff_filter_frame(outlink, av_frame_clone(pic));
761  if (ret < 0)
762  return ret;
763  }
764 
765  if (ebur128->metadata) { /* happens only once per filter_frame call */
766  char metabuf[128];
767 #define META_PREFIX "lavfi.r128."
768 
769 #define SET_META(name, var) do { \
770  snprintf(metabuf, sizeof(metabuf), "%.3f", var); \
771  av_dict_set(&insamples->metadata, name, metabuf, 0); \
772 } while (0)
773 
774 #define SET_META_PEAK(name, ptype) do { \
775  if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
776  char key[64]; \
777  for (ch = 0; ch < nb_channels; ch++) { \
778  snprintf(key, sizeof(key), \
779  META_PREFIX AV_STRINGIFY(name) "_peaks_ch%d", ch); \
780  SET_META(key, ebur128->name##_peaks[ch]); \
781  } \
782  } \
783 } while (0)
784 
785  SET_META(META_PREFIX "M", loudness_400);
786  SET_META(META_PREFIX "S", loudness_3000);
788  SET_META(META_PREFIX "LRA", ebur128->loudness_range);
789  SET_META(META_PREFIX "LRA.low", ebur128->lra_low);
790  SET_META(META_PREFIX "LRA.high", ebur128->lra_high);
791 
793  SET_META_PEAK(true, TRUE);
794  }
795 
796  av_log(ctx, ebur128->loglevel, "t: %-10s " LOG_FMT,
797  av_ts2timestr(pts, &outlink->time_base),
798  loudness_400, loudness_3000,
799  ebur128->integrated_loudness, ebur128->loudness_range);
800 
801 #define PRINT_PEAKS(str, sp, ptype) do { \
802  if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
803  av_log(ctx, ebur128->loglevel, " " str ":"); \
804  for (ch = 0; ch < nb_channels; ch++) \
805  av_log(ctx, ebur128->loglevel, " %5.1f", DBFS(sp[ch])); \
806  av_log(ctx, ebur128->loglevel, " dBFS"); \
807  } \
808 } while (0)
809 
810  PRINT_PEAKS("SPK", ebur128->sample_peaks, SAMPLES);
811  PRINT_PEAKS("FTPK", ebur128->true_peaks_per_frame, TRUE);
812  PRINT_PEAKS("TPK", ebur128->true_peaks, TRUE);
813  av_log(ctx, ebur128->loglevel, "\n");
814  }
815  }
816 
817  return ff_filter_frame(ctx->outputs[ebur128->do_video], insamples);
818 }
819 
821 {
822  EBUR128Context *ebur128 = ctx->priv;
825  AVFilterLink *inlink = ctx->inputs[0];
826  AVFilterLink *outlink = ctx->outputs[0];
827 
829  static const int input_srate[] = {48000, -1}; // ITU-R BS.1770 provides coeff only for 48kHz
830  static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_RGB24, AV_PIX_FMT_NONE };
831 
832  /* set optional output video format */
833  if (ebur128->do_video) {
834  formats = ff_make_format_list(pix_fmts);
835  if (!formats)
836  return AVERROR(ENOMEM);
837  ff_formats_ref(formats, &outlink->in_formats);
838  outlink = ctx->outputs[1];
839  }
840 
841  /* set input and output audio formats
842  * Note: ff_set_common_* functions are not used because they affect all the
843  * links, and thus break the video format negotiation */
844  formats = ff_make_format_list(sample_fmts);
845  if (!formats)
846  return AVERROR(ENOMEM);
847  ff_formats_ref(formats, &inlink->out_formats);
848  ff_formats_ref(formats, &outlink->in_formats);
849 
850  layouts = ff_all_channel_layouts();
851  if (!layouts)
852  return AVERROR(ENOMEM);
853  ff_channel_layouts_ref(layouts, &inlink->out_channel_layouts);
854  ff_channel_layouts_ref(layouts, &outlink->in_channel_layouts);
855 
856  formats = ff_make_format_list(input_srate);
857  if (!formats)
858  return AVERROR(ENOMEM);
859  ff_formats_ref(formats, &inlink->out_samplerates);
860  ff_formats_ref(formats, &outlink->in_samplerates);
861 
862  return 0;
863 }
864 
865 static av_cold void uninit(AVFilterContext *ctx)
866 {
867  int i;
868  EBUR128Context *ebur128 = ctx->priv;
869 
870  av_log(ctx, AV_LOG_INFO, "Summary:\n\n"
871  " Integrated loudness:\n"
872  " I: %5.1f LUFS\n"
873  " Threshold: %5.1f LUFS\n\n"
874  " Loudness range:\n"
875  " LRA: %5.1f LU\n"
876  " Threshold: %5.1f LUFS\n"
877  " LRA low: %5.1f LUFS\n"
878  " LRA high: %5.1f LUFS",
879  ebur128->integrated_loudness, ebur128->i400.rel_threshold,
880  ebur128->loudness_range, ebur128->i3000.rel_threshold,
881  ebur128->lra_low, ebur128->lra_high);
882 
883 #define PRINT_PEAK_SUMMARY(str, sp, ptype) do { \
884  int ch; \
885  double maxpeak; \
886  maxpeak = 0.0; \
887  if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
888  for (ch = 0; ch < ebur128->nb_channels; ch++) \
889  maxpeak = FFMAX(maxpeak, sp[ch]); \
890  av_log(ctx, AV_LOG_INFO, "\n\n " str " peak:\n" \
891  " Peak: %5.1f dBFS", \
892  DBFS(maxpeak)); \
893  } \
894 } while (0)
895 
896  PRINT_PEAK_SUMMARY("Sample", ebur128->sample_peaks, SAMPLES);
897  PRINT_PEAK_SUMMARY("True", ebur128->true_peaks, TRUE);
898  av_log(ctx, AV_LOG_INFO, "\n");
899 
900  av_freep(&ebur128->y_line_ref);
901  av_freep(&ebur128->ch_weighting);
902  av_freep(&ebur128->true_peaks);
903  av_freep(&ebur128->sample_peaks);
904  av_freep(&ebur128->true_peaks_per_frame);
905  av_freep(&ebur128->i400.histogram);
906  av_freep(&ebur128->i3000.histogram);
907  for (i = 0; i < ebur128->nb_channels; i++) {
908  av_freep(&ebur128->i400.cache[i]);
909  av_freep(&ebur128->i3000.cache[i]);
910  }
911  for (i = 0; i < ctx->nb_outputs; i++)
912  av_freep(&ctx->output_pads[i].name);
913  av_frame_free(&ebur128->outpicref);
914 #if CONFIG_SWRESAMPLE
915  av_freep(&ebur128->swr_buf);
916  swr_free(&ebur128->swr_ctx);
917 #endif
918 }
919 
920 static const AVFilterPad ebur128_inputs[] = {
921  {
922  .name = "default",
923  .type = AVMEDIA_TYPE_AUDIO,
924  .filter_frame = filter_frame,
925  .config_props = config_audio_input,
926  },
927  { NULL }
928 };
929 
931  .name = "ebur128",
932  .description = NULL_IF_CONFIG_SMALL("EBU R128 scanner."),
933  .priv_size = sizeof(EBUR128Context),
934  .init = init,
935  .uninit = uninit,
937  .inputs = ebur128_inputs,
938  .outputs = NULL,
939  .priv_class = &ebur128_class,
941 };