summaryrefslogtreecommitdiff
path: root/libavfilter/f_ebur128.c (plain)
blob: 1fd85bbf790073833eade3d453efc1a98b0a1f4d
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
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21/**
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/ffmath.h"
37#include "libavutil/xga_font_data.h"
38#include "libavutil/opt.h"
39#include "libavutil/timestamp.h"
40#include "libswresample/swresample.h"
41#include "audio.h"
42#include "avfilter.h"
43#include "formats.h"
44#include "internal.h"
45
46#define MAX_CHANNELS 63
47
48/* pre-filter coefficients */
49#define PRE_B0 1.53512485958697
50#define PRE_B1 -2.69169618940638
51#define PRE_B2 1.19839281085285
52#define PRE_A1 -1.69065929318241
53#define PRE_A2 0.73248077421585
54
55/* RLB-filter coefficients */
56#define RLB_B0 1.0
57#define RLB_B1 -2.0
58#define RLB_B2 1.0
59#define RLB_A1 -1.99004745483398
60#define RLB_A2 0.99007225036621
61
62#define ABS_THRES -70 ///< silence gate: we discard anything below this absolute (LUFS) threshold
63#define ABS_UP_THRES 10 ///< upper loud limit to consider (ABS_THRES being the minimum)
64#define HIST_GRAIN 100 ///< defines histogram precision
65#define HIST_SIZE ((ABS_UP_THRES - ABS_THRES) * HIST_GRAIN + 1)
66
67/**
68 * A histogram is an array of HIST_SIZE hist_entry storing all the energies
69 * recorded (with an accuracy of 1/HIST_GRAIN) of the loudnesses from ABS_THRES
70 * (at 0) to ABS_UP_THRES (at HIST_SIZE-1).
71 * This fixed-size system avoids the need of a list of energies growing
72 * infinitely over the time and is thus more scalable.
73 */
74struct hist_entry {
75 int count; ///< how many times the corresponding value occurred
76 double energy; ///< E = 10^((L + 0.691) / 10)
77 double loudness; ///< L = -0.691 + 10 * log10(E)
78};
79
80struct integrator {
81 double *cache[MAX_CHANNELS]; ///< window of filtered samples (N ms)
82 int cache_pos; ///< focus on the last added bin in the cache array
83 double sum[MAX_CHANNELS]; ///< sum of the last N ms filtered samples (cache content)
84 int filled; ///< 1 if the cache is completely filled, 0 otherwise
85 double rel_threshold; ///< relative threshold
86 double sum_kept_powers; ///< sum of the powers (weighted sums) above absolute threshold
87 int nb_kept_powers; ///< number of sum above absolute threshold
88 struct hist_entry *histogram; ///< histogram of the powers, used to compute LRA and I
89};
90
91struct rect { int x, y, w, h; };
92
93typedef struct {
94 const AVClass *class; ///< AVClass context for log and options purpose
95
96 /* peak metering */
97 int peak_mode; ///< enabled peak modes
98 double *true_peaks; ///< true peaks per channel
99 double *sample_peaks; ///< sample peaks per channel
100 double *true_peaks_per_frame; ///< true peaks in a frame per channel
101#if CONFIG_SWRESAMPLE
102 SwrContext *swr_ctx; ///< over-sampling context for true peak metering
103 double *swr_buf; ///< resampled audio data for true peak metering
104 int swr_linesize;
105#endif
106
107 /* video */
108 int do_video; ///< 1 if video output enabled, 0 otherwise
109 int w, h; ///< size of the video output
110 struct rect text; ///< rectangle for the LU legend on the left
111 struct rect graph; ///< rectangle for the main graph in the center
112 struct rect gauge; ///< rectangle for the gauge on the right
113 AVFrame *outpicref; ///< output picture reference, updated regularly
114 int meter; ///< select a EBU mode between +9 and +18
115 int scale_range; ///< the range of LU values according to the meter
116 int y_zero_lu; ///< the y value (pixel position) for 0 LU
117 int *y_line_ref; ///< y reference values for drawing the LU lines in the graph and the gauge
118
119 /* audio */
120 int nb_channels; ///< number of channels in the input
121 double *ch_weighting; ///< channel weighting mapping
122 int sample_count; ///< sample count used for refresh frequency, reset at refresh
123
124 /* Filter caches.
125 * The mult by 3 in the following is for X[i], X[i-1] and X[i-2] */
126 double x[MAX_CHANNELS * 3]; ///< 3 input samples cache for each channel
127 double y[MAX_CHANNELS * 3]; ///< 3 pre-filter samples cache for each channel
128 double z[MAX_CHANNELS * 3]; ///< 3 RLB-filter samples cache for each channel
129
130#define I400_BINS (48000 * 4 / 10)
131#define I3000_BINS (48000 * 3)
132 struct integrator i400; ///< 400ms integrator, used for Momentary loudness (M), and Integrated loudness (I)
133 struct integrator i3000; ///< 3s integrator, used for Short term loudness (S), and Loudness Range (LRA)
134
135 /* I and LRA specific */
136 double integrated_loudness; ///< integrated loudness in LUFS (I)
137 double loudness_range; ///< loudness range in LU (LRA)
138 double lra_low, lra_high; ///< low and high LRA values
139
140 /* misc */
141 int loglevel; ///< log level for frame logging
142 int metadata; ///< whether or not to inject loudness results in frames
143 int dual_mono; ///< whether or not to treat single channel input files as dual-mono
144 double pan_law; ///< pan law value used to calculate dual-mono measurements
145} EBUR128Context;
146
147enum {
148 PEAK_MODE_NONE = 0,
149 PEAK_MODE_SAMPLES_PEAKS = 1<<1,
150 PEAK_MODE_TRUE_PEAKS = 1<<2,
151};
152
153#define OFFSET(x) offsetof(EBUR128Context, x)
154#define A AV_OPT_FLAG_AUDIO_PARAM
155#define V AV_OPT_FLAG_VIDEO_PARAM
156#define F AV_OPT_FLAG_FILTERING_PARAM
157static const AVOption ebur128_options[] = {
158 { "video", "set video output", OFFSET(do_video), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, V|F },
159 { "size", "set video size", OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "640x480"}, 0, 0, V|F },
160 { "meter", "set scale meter (+9 to +18)", OFFSET(meter), AV_OPT_TYPE_INT, {.i64 = 9}, 9, 18, V|F },
161 { "framelog", "force frame logging level", OFFSET(loglevel), AV_OPT_TYPE_INT, {.i64 = -1}, INT_MIN, INT_MAX, A|V|F, "level" },
162 { "info", "information logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_INFO}, INT_MIN, INT_MAX, A|V|F, "level" },
163 { "verbose", "verbose logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_VERBOSE}, INT_MIN, INT_MAX, A|V|F, "level" },
164 { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, A|V|F },
165 { "peak", "set peak mode", OFFSET(peak_mode), AV_OPT_TYPE_FLAGS, {.i64 = PEAK_MODE_NONE}, 0, INT_MAX, A|F, "mode" },
166 { "none", "disable any peak mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_NONE}, INT_MIN, INT_MAX, A|F, "mode" },
167 { "sample", "enable peak-sample mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_SAMPLES_PEAKS}, INT_MIN, INT_MAX, A|F, "mode" },
168 { "true", "enable true-peak mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_TRUE_PEAKS}, INT_MIN, INT_MAX, A|F, "mode" },
169 { "dualmono", "treat mono input files as dual-mono", OFFSET(dual_mono), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, A|F },
170 { "panlaw", "set a specific pan law for dual-mono files", OFFSET(pan_law), AV_OPT_TYPE_DOUBLE, {.dbl = -3.01029995663978}, -10.0, 0.0, A|F },
171 { NULL },
172};
173
174AVFILTER_DEFINE_CLASS(ebur128);
175
176static const uint8_t graph_colors[] = {
177 0xdd, 0x66, 0x66, // value above 0LU non reached
178 0x66, 0x66, 0xdd, // value below 0LU non reached
179 0x96, 0x33, 0x33, // value above 0LU reached
180 0x33, 0x33, 0x96, // value below 0LU reached
181 0xdd, 0x96, 0x96, // value above 0LU line non reached
182 0x96, 0x96, 0xdd, // value below 0LU line non reached
183 0xdd, 0x33, 0x33, // value above 0LU line reached
184 0x33, 0x33, 0xdd, // value below 0LU line reached
185};
186
187static const uint8_t *get_graph_color(const EBUR128Context *ebur128, int v, int y)
188{
189 const int below0 = y > ebur128->y_zero_lu;
190 const int reached = y >= v;
191 const int line = ebur128->y_line_ref[y] || y == ebur128->y_zero_lu;
192 const int colorid = 4*line + 2*reached + below0;
193 return graph_colors + 3*colorid;
194}
195
196static inline int lu_to_y(const EBUR128Context *ebur128, double v)
197{
198 v += 2 * ebur128->meter; // make it in range [0;...]
199 v = av_clipf(v, 0, ebur128->scale_range); // make sure it's in the graph scale
200 v = ebur128->scale_range - v; // invert value (y=0 is on top)
201 return v * ebur128->graph.h / ebur128->scale_range; // rescale from scale range to px height
202}
203
204#define FONT8 0
205#define FONT16 1
206
207static const uint8_t font_colors[] = {
208 0xdd, 0xdd, 0x00,
209 0x00, 0x96, 0x96,
210};
211
212static void drawtext(AVFrame *pic, int x, int y, int ftid, const uint8_t *color, const char *fmt, ...)
213{
214 int i;
215 char buf[128] = {0};
216 const uint8_t *font;
217 int font_height;
218 va_list vl;
219
220 if (ftid == FONT16) font = avpriv_vga16_font, font_height = 16;
221 else if (ftid == FONT8) font = avpriv_cga_font, font_height = 8;
222 else return;
223
224 va_start(vl, fmt);
225 vsnprintf(buf, sizeof(buf), fmt, vl);
226 va_end(vl);
227
228 for (i = 0; buf[i]; i++) {
229 int char_y, mask;
230 uint8_t *p = pic->data[0] + y*pic->linesize[0] + (x + i*8)*3;
231
232 for (char_y = 0; char_y < font_height; char_y++) {
233 for (mask = 0x80; mask; mask >>= 1) {
234 if (font[buf[i] * font_height + char_y] & mask)
235 memcpy(p, color, 3);
236 else
237 memcpy(p, "\x00\x00\x00", 3);
238 p += 3;
239 }
240 p += pic->linesize[0] - 8*3;
241 }
242 }
243}
244
245static void drawline(AVFrame *pic, int x, int y, int len, int step)
246{
247 int i;
248 uint8_t *p = pic->data[0] + y*pic->linesize[0] + x*3;
249
250 for (i = 0; i < len; i++) {
251 memcpy(p, "\x00\xff\x00", 3);
252 p += step;
253 }
254}
255
256static int config_video_output(AVFilterLink *outlink)
257{
258 int i, x, y;
259 uint8_t *p;
260 AVFilterContext *ctx = outlink->src;
261 EBUR128Context *ebur128 = ctx->priv;
262 AVFrame *outpicref;
263
264 /* check if there is enough space to represent everything decently */
265 if (ebur128->w < 640 || ebur128->h < 480) {
266 av_log(ctx, AV_LOG_ERROR, "Video size %dx%d is too small, "
267 "minimum size is 640x480\n", ebur128->w, ebur128->h);
268 return AVERROR(EINVAL);
269 }
270 outlink->w = ebur128->w;
271 outlink->h = ebur128->h;
272
273#define PAD 8
274
275 /* configure text area position and size */
276 ebur128->text.x = PAD;
277 ebur128->text.y = 40;
278 ebur128->text.w = 3 * 8; // 3 characters
279 ebur128->text.h = ebur128->h - PAD - ebur128->text.y;
280
281 /* configure gauge position and size */
282 ebur128->gauge.w = 20;
283 ebur128->gauge.h = ebur128->text.h;
284 ebur128->gauge.x = ebur128->w - PAD - ebur128->gauge.w;
285 ebur128->gauge.y = ebur128->text.y;
286
287 /* configure graph position and size */
288 ebur128->graph.x = ebur128->text.x + ebur128->text.w + PAD;
289 ebur128->graph.y = ebur128->gauge.y;
290 ebur128->graph.w = ebur128->gauge.x - ebur128->graph.x - PAD;
291 ebur128->graph.h = ebur128->gauge.h;
292
293 /* graph and gauge share the LU-to-pixel code */
294 av_assert0(ebur128->graph.h == ebur128->gauge.h);
295
296 /* prepare the initial picref buffer */
297 av_frame_free(&ebur128->outpicref);
298 ebur128->outpicref = outpicref =
299 ff_get_video_buffer(outlink, outlink->w, outlink->h);
300 if (!outpicref)
301 return AVERROR(ENOMEM);
302 outlink->sample_aspect_ratio = (AVRational){1,1};
303
304 /* init y references values (to draw LU lines) */
305 ebur128->y_line_ref = av_calloc(ebur128->graph.h + 1, sizeof(*ebur128->y_line_ref));
306 if (!ebur128->y_line_ref)
307 return AVERROR(ENOMEM);
308
309 /* black background */
310 memset(outpicref->data[0], 0, ebur128->h * outpicref->linesize[0]);
311
312 /* draw LU legends */
313 drawtext(outpicref, PAD, PAD+16, FONT8, font_colors+3, " LU");
314 for (i = ebur128->meter; i >= -ebur128->meter * 2; i--) {
315 y = lu_to_y(ebur128, i);
316 x = PAD + (i < 10 && i > -10) * 8;
317 ebur128->y_line_ref[y] = i;
318 y -= 4; // -4 to center vertically
319 drawtext(outpicref, x, y + ebur128->graph.y, FONT8, font_colors+3,
320 "%c%d", i < 0 ? '-' : i > 0 ? '+' : ' ', FFABS(i));
321 }
322
323 /* draw graph */
324 ebur128->y_zero_lu = lu_to_y(ebur128, 0);
325 p = outpicref->data[0] + ebur128->graph.y * outpicref->linesize[0]
326 + ebur128->graph.x * 3;
327 for (y = 0; y < ebur128->graph.h; y++) {
328 const uint8_t *c = get_graph_color(ebur128, INT_MAX, y);
329
330 for (x = 0; x < ebur128->graph.w; x++)
331 memcpy(p + x*3, c, 3);
332 p += outpicref->linesize[0];
333 }
334
335 /* draw fancy rectangles around the graph and the gauge */
336#define DRAW_RECT(r) do { \
337 drawline(outpicref, r.x, r.y - 1, r.w, 3); \
338 drawline(outpicref, r.x, r.y + r.h, r.w, 3); \
339 drawline(outpicref, r.x - 1, r.y, r.h, outpicref->linesize[0]); \
340 drawline(outpicref, r.x + r.w, r.y, r.h, outpicref->linesize[0]); \
341} while (0)
342 DRAW_RECT(ebur128->graph);
343 DRAW_RECT(ebur128->gauge);
344
345 return 0;
346}
347
348static int config_audio_input(AVFilterLink *inlink)
349{
350 AVFilterContext *ctx = inlink->dst;
351 EBUR128Context *ebur128 = ctx->priv;
352
353 /* Force 100ms framing in case of metadata injection: the frames must have
354 * a granularity of the window overlap to be accurately exploited.
355 * As for the true peaks mode, it just simplifies the resampling buffer
356 * allocation and the lookup in it (since sample buffers differ in size, it
357 * can be more complex to integrate in the one-sample loop of
358 * filter_frame()). */
359 if (ebur128->metadata || (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS))
360 inlink->min_samples =
361 inlink->max_samples =
362 inlink->partial_buf_size = inlink->sample_rate / 10;
363 return 0;
364}
365
366static int config_audio_output(AVFilterLink *outlink)
367{
368 int i;
369 AVFilterContext *ctx = outlink->src;
370 EBUR128Context *ebur128 = ctx->priv;
371 const int nb_channels = av_get_channel_layout_nb_channels(outlink->channel_layout);
372
373#define BACK_MASK (AV_CH_BACK_LEFT |AV_CH_BACK_CENTER |AV_CH_BACK_RIGHT| \
374 AV_CH_TOP_BACK_LEFT|AV_CH_TOP_BACK_CENTER|AV_CH_TOP_BACK_RIGHT| \
375 AV_CH_SIDE_LEFT |AV_CH_SIDE_RIGHT| \
376 AV_CH_SURROUND_DIRECT_LEFT |AV_CH_SURROUND_DIRECT_RIGHT)
377
378 ebur128->nb_channels = nb_channels;
379 ebur128->ch_weighting = av_calloc(nb_channels, sizeof(*ebur128->ch_weighting));
380 if (!ebur128->ch_weighting)
381 return AVERROR(ENOMEM);
382
383 for (i = 0; i < nb_channels; i++) {
384 /* channel weighting */
385 const uint16_t chl = av_channel_layout_extract_channel(outlink->channel_layout, i);
386 if (chl & (AV_CH_LOW_FREQUENCY|AV_CH_LOW_FREQUENCY_2)) {
387 ebur128->ch_weighting[i] = 0;
388 } else if (chl & BACK_MASK) {
389 ebur128->ch_weighting[i] = 1.41;
390 } else {
391 ebur128->ch_weighting[i] = 1.0;
392 }
393
394 if (!ebur128->ch_weighting[i])
395 continue;
396
397 /* bins buffer for the two integration window (400ms and 3s) */
398 ebur128->i400.cache[i] = av_calloc(I400_BINS, sizeof(*ebur128->i400.cache[0]));
399 ebur128->i3000.cache[i] = av_calloc(I3000_BINS, sizeof(*ebur128->i3000.cache[0]));
400 if (!ebur128->i400.cache[i] || !ebur128->i3000.cache[i])
401 return AVERROR(ENOMEM);
402 }
403
404#if CONFIG_SWRESAMPLE
405 if (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS) {
406 int ret;
407
408 ebur128->swr_buf = av_malloc_array(nb_channels, 19200 * sizeof(double));
409 ebur128->true_peaks = av_calloc(nb_channels, sizeof(*ebur128->true_peaks));
410 ebur128->true_peaks_per_frame = av_calloc(nb_channels, sizeof(*ebur128->true_peaks_per_frame));
411 ebur128->swr_ctx = swr_alloc();
412 if (!ebur128->swr_buf || !ebur128->true_peaks ||
413 !ebur128->true_peaks_per_frame || !ebur128->swr_ctx)
414 return AVERROR(ENOMEM);
415
416 av_opt_set_int(ebur128->swr_ctx, "in_channel_layout", outlink->channel_layout, 0);
417 av_opt_set_int(ebur128->swr_ctx, "in_sample_rate", outlink->sample_rate, 0);
418 av_opt_set_sample_fmt(ebur128->swr_ctx, "in_sample_fmt", outlink->format, 0);
419
420 av_opt_set_int(ebur128->swr_ctx, "out_channel_layout", outlink->channel_layout, 0);
421 av_opt_set_int(ebur128->swr_ctx, "out_sample_rate", 192000, 0);
422 av_opt_set_sample_fmt(ebur128->swr_ctx, "out_sample_fmt", outlink->format, 0);
423
424 ret = swr_init(ebur128->swr_ctx);
425 if (ret < 0)
426 return ret;
427 }
428#endif
429
430 if (ebur128->peak_mode & PEAK_MODE_SAMPLES_PEAKS) {
431 ebur128->sample_peaks = av_calloc(nb_channels, sizeof(*ebur128->sample_peaks));
432 if (!ebur128->sample_peaks)
433 return AVERROR(ENOMEM);
434 }
435
436 return 0;
437}
438
439#define ENERGY(loudness) (ff_exp10(((loudness) + 0.691) / 10.))
440#define LOUDNESS(energy) (-0.691 + 10 * log10(energy))
441#define DBFS(energy) (20 * log10(energy))
442
443static struct hist_entry *get_histogram(void)
444{
445 int i;
446 struct hist_entry *h = av_calloc(HIST_SIZE, sizeof(*h));
447
448 if (!h)
449 return NULL;
450 for (i = 0; i < HIST_SIZE; i++) {
451 h[i].loudness = i / (double)HIST_GRAIN + ABS_THRES;
452 h[i].energy = ENERGY(h[i].loudness);
453 }
454 return h;
455}
456
457static av_cold int init(AVFilterContext *ctx)
458{
459 EBUR128Context *ebur128 = ctx->priv;
460 AVFilterPad pad;
461
462 if (ebur128->loglevel != AV_LOG_INFO &&
463 ebur128->loglevel != AV_LOG_VERBOSE) {
464 if (ebur128->do_video || ebur128->metadata)
465 ebur128->loglevel = AV_LOG_VERBOSE;
466 else
467 ebur128->loglevel = AV_LOG_INFO;
468 }
469
470 if (!CONFIG_SWRESAMPLE && (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS)) {
471 av_log(ctx, AV_LOG_ERROR,
472 "True-peak mode requires libswresample to be performed\n");
473 return AVERROR(EINVAL);
474 }
475
476 // if meter is +9 scale, scale range is from -18 LU to +9 LU (or 3*9)
477 // if meter is +18 scale, scale range is from -36 LU to +18 LU (or 3*18)
478 ebur128->scale_range = 3 * ebur128->meter;
479
480 ebur128->i400.histogram = get_histogram();
481 ebur128->i3000.histogram = get_histogram();
482 if (!ebur128->i400.histogram || !ebur128->i3000.histogram)
483 return AVERROR(ENOMEM);
484
485 ebur128->integrated_loudness = ABS_THRES;
486 ebur128->loudness_range = 0;
487
488 /* insert output pads */
489 if (ebur128->do_video) {
490 pad = (AVFilterPad){
491 .name = av_strdup("out0"),
492 .type = AVMEDIA_TYPE_VIDEO,
493 .config_props = config_video_output,
494 };
495 if (!pad.name)
496 return AVERROR(ENOMEM);
497 ff_insert_outpad(ctx, 0, &pad);
498 }
499 pad = (AVFilterPad){
500 .name = av_asprintf("out%d", ebur128->do_video),
501 .type = AVMEDIA_TYPE_AUDIO,
502 .config_props = config_audio_output,
503 };
504 if (!pad.name)
505 return AVERROR(ENOMEM);
506 ff_insert_outpad(ctx, ebur128->do_video, &pad);
507
508 /* summary */
509 av_log(ctx, AV_LOG_VERBOSE, "EBU +%d scale\n", ebur128->meter);
510
511 return 0;
512}
513
514#define HIST_POS(power) (int)(((power) - ABS_THRES) * HIST_GRAIN)
515
516/* loudness and power should be set such as loudness = -0.691 +
517 * 10*log10(power), we just avoid doing that calculus two times */
518static int gate_update(struct integrator *integ, double power,
519 double loudness, int gate_thres)
520{
521 int ipower;
522 double relative_threshold;
523 int gate_hist_pos;
524
525 /* update powers histograms by incrementing current power count */
526 ipower = av_clip(HIST_POS(loudness), 0, HIST_SIZE - 1);
527 integ->histogram[ipower].count++;
528
529 /* compute relative threshold and get its position in the histogram */
530 integ->sum_kept_powers += power;
531 integ->nb_kept_powers++;
532 relative_threshold = integ->sum_kept_powers / integ->nb_kept_powers;
533 if (!relative_threshold)
534 relative_threshold = 1e-12;
535 integ->rel_threshold = LOUDNESS(relative_threshold) + gate_thres;
536 gate_hist_pos = av_clip(HIST_POS(integ->rel_threshold), 0, HIST_SIZE - 1);
537
538 return gate_hist_pos;
539}
540
541static int filter_frame(AVFilterLink *inlink, AVFrame *insamples)
542{
543 int i, ch, idx_insample;
544 AVFilterContext *ctx = inlink->dst;
545 EBUR128Context *ebur128 = ctx->priv;
546 const int nb_channels = ebur128->nb_channels;
547 const int nb_samples = insamples->nb_samples;
548 const double *samples = (double *)insamples->data[0];
549 AVFrame *pic = ebur128->outpicref;
550
551#if CONFIG_SWRESAMPLE
552 if (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS) {
553 const double *swr_samples = ebur128->swr_buf;
554 int ret = swr_convert(ebur128->swr_ctx, (uint8_t**)&ebur128->swr_buf, 19200,
555 (const uint8_t **)insamples->data, nb_samples);
556 if (ret < 0)
557 return ret;
558 for (ch = 0; ch < nb_channels; ch++)
559 ebur128->true_peaks_per_frame[ch] = 0.0;
560 for (idx_insample = 0; idx_insample < ret; idx_insample++) {
561 for (ch = 0; ch < nb_channels; ch++) {
562 ebur128->true_peaks[ch] = FFMAX(ebur128->true_peaks[ch], fabs(*swr_samples));
563 ebur128->true_peaks_per_frame[ch] = FFMAX(ebur128->true_peaks_per_frame[ch],
564 fabs(*swr_samples));
565 swr_samples++;
566 }
567 }
568 }
569#endif
570
571 for (idx_insample = 0; idx_insample < nb_samples; idx_insample++) {
572 const int bin_id_400 = ebur128->i400.cache_pos;
573 const int bin_id_3000 = ebur128->i3000.cache_pos;
574
575#define MOVE_TO_NEXT_CACHED_ENTRY(time) do { \
576 ebur128->i##time.cache_pos++; \
577 if (ebur128->i##time.cache_pos == I##time##_BINS) { \
578 ebur128->i##time.filled = 1; \
579 ebur128->i##time.cache_pos = 0; \
580 } \
581} while (0)
582
583 MOVE_TO_NEXT_CACHED_ENTRY(400);
584 MOVE_TO_NEXT_CACHED_ENTRY(3000);
585
586 for (ch = 0; ch < nb_channels; ch++) {
587 double bin;
588
589 if (ebur128->peak_mode & PEAK_MODE_SAMPLES_PEAKS)
590 ebur128->sample_peaks[ch] = FFMAX(ebur128->sample_peaks[ch], fabs(*samples));
591
592 ebur128->x[ch * 3] = *samples++; // set X[i]
593
594 if (!ebur128->ch_weighting[ch])
595 continue;
596
597 /* Y[i] = X[i]*b0 + X[i-1]*b1 + X[i-2]*b2 - Y[i-1]*a1 - Y[i-2]*a2 */
598#define FILTER(Y, X, name) do { \
599 double *dst = ebur128->Y + ch*3; \
600 double *src = ebur128->X + ch*3; \
601 dst[2] = dst[1]; \
602 dst[1] = dst[0]; \
603 dst[0] = src[0]*name##_B0 + src[1]*name##_B1 + src[2]*name##_B2 \
604 - dst[1]*name##_A1 - dst[2]*name##_A2; \
605} while (0)
606
607 // TODO: merge both filters in one?
608 FILTER(y, x, PRE); // apply pre-filter
609 ebur128->x[ch * 3 + 2] = ebur128->x[ch * 3 + 1];
610 ebur128->x[ch * 3 + 1] = ebur128->x[ch * 3 ];
611 FILTER(z, y, RLB); // apply RLB-filter
612
613 bin = ebur128->z[ch * 3] * ebur128->z[ch * 3];
614
615 /* add the new value, and limit the sum to the cache size (400ms or 3s)
616 * by removing the oldest one */
617 ebur128->i400.sum [ch] = ebur128->i400.sum [ch] + bin - ebur128->i400.cache [ch][bin_id_400];
618 ebur128->i3000.sum[ch] = ebur128->i3000.sum[ch] + bin - ebur128->i3000.cache[ch][bin_id_3000];
619
620 /* override old cache entry with the new value */
621 ebur128->i400.cache [ch][bin_id_400 ] = bin;
622 ebur128->i3000.cache[ch][bin_id_3000] = bin;
623 }
624
625 /* For integrated loudness, gating blocks are 400ms long with 75%
626 * overlap (see BS.1770-2 p5), so a re-computation is needed each 100ms
627 * (4800 samples at 48kHz). */
628 if (++ebur128->sample_count == 4800) {
629 double loudness_400, loudness_3000;
630 double power_400 = 1e-12, power_3000 = 1e-12;
631 AVFilterLink *outlink = ctx->outputs[0];
632 const int64_t pts = insamples->pts +
633 av_rescale_q(idx_insample, (AVRational){ 1, inlink->sample_rate },
634 outlink->time_base);
635
636 ebur128->sample_count = 0;
637
638#define COMPUTE_LOUDNESS(m, time) do { \
639 if (ebur128->i##time.filled) { \
640 /* weighting sum of the last <time> ms */ \
641 for (ch = 0; ch < nb_channels; ch++) \
642 power_##time += ebur128->ch_weighting[ch] * ebur128->i##time.sum[ch]; \
643 power_##time /= I##time##_BINS; \
644 } \
645 loudness_##time = LOUDNESS(power_##time); \
646} while (0)
647
648 COMPUTE_LOUDNESS(M, 400);
649 COMPUTE_LOUDNESS(S, 3000);
650
651 /* Integrated loudness */
652#define I_GATE_THRES -10 // initially defined to -8 LU in the first EBU standard
653
654 if (loudness_400 >= ABS_THRES) {
655 double integrated_sum = 0;
656 int nb_integrated = 0;
657 int gate_hist_pos = gate_update(&ebur128->i400, power_400,
658 loudness_400, I_GATE_THRES);
659
660 /* compute integrated loudness by summing the histogram values
661 * above the relative threshold */
662 for (i = gate_hist_pos; i < HIST_SIZE; i++) {
663 const int nb_v = ebur128->i400.histogram[i].count;
664 nb_integrated += nb_v;
665 integrated_sum += nb_v * ebur128->i400.histogram[i].energy;
666 }
667 if (nb_integrated) {
668 ebur128->integrated_loudness = LOUDNESS(integrated_sum / nb_integrated);
669 /* dual-mono correction */
670 if (nb_channels == 1 && ebur128->dual_mono) {
671 ebur128->integrated_loudness -= ebur128->pan_law;
672 }
673 }
674 }
675
676 /* LRA */
677#define LRA_GATE_THRES -20
678#define LRA_LOWER_PRC 10
679#define LRA_HIGHER_PRC 95
680
681 /* XXX: example code in EBU 3342 is ">=" but formula in BS.1770
682 * specs is ">" */
683 if (loudness_3000 >= ABS_THRES) {
684 int nb_powers = 0;
685 int gate_hist_pos = gate_update(&ebur128->i3000, power_3000,
686 loudness_3000, LRA_GATE_THRES);
687
688 for (i = gate_hist_pos; i < HIST_SIZE; i++)
689 nb_powers += ebur128->i3000.histogram[i].count;
690 if (nb_powers) {
691 int n, nb_pow;
692
693 /* get lower loudness to consider */
694 n = 0;
695 nb_pow = LRA_LOWER_PRC * nb_powers / 100. + 0.5;
696 for (i = gate_hist_pos; i < HIST_SIZE; i++) {
697 n += ebur128->i3000.histogram[i].count;
698 if (n >= nb_pow) {
699 ebur128->lra_low = ebur128->i3000.histogram[i].loudness;
700 break;
701 }
702 }
703
704 /* get higher loudness to consider */
705 n = nb_powers;
706 nb_pow = LRA_HIGHER_PRC * nb_powers / 100. + 0.5;
707 for (i = HIST_SIZE - 1; i >= 0; i--) {
708 n -= ebur128->i3000.histogram[i].count;
709 if (n < nb_pow) {
710 ebur128->lra_high = ebur128->i3000.histogram[i].loudness;
711 break;
712 }
713 }
714
715 // XXX: show low & high on the graph?
716 ebur128->loudness_range = ebur128->lra_high - ebur128->lra_low;
717 }
718 }
719
720 /* dual-mono correction */
721 if (nb_channels == 1 && ebur128->dual_mono) {
722 loudness_400 -= ebur128->pan_law;
723 loudness_3000 -= ebur128->pan_law;
724 }
725
726#define LOG_FMT "M:%6.1f S:%6.1f I:%6.1f LUFS LRA:%6.1f LU"
727
728 /* push one video frame */
729 if (ebur128->do_video) {
730 int x, y, ret;
731 uint8_t *p;
732
733 const int y_loudness_lu_graph = lu_to_y(ebur128, loudness_3000 + 23);
734 const int y_loudness_lu_gauge = lu_to_y(ebur128, loudness_400 + 23);
735
736 /* draw the graph using the short-term loudness */
737 p = pic->data[0] + ebur128->graph.y*pic->linesize[0] + ebur128->graph.x*3;
738 for (y = 0; y < ebur128->graph.h; y++) {
739 const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_graph, y);
740
741 memmove(p, p + 3, (ebur128->graph.w - 1) * 3);
742 memcpy(p + (ebur128->graph.w - 1) * 3, c, 3);
743 p += pic->linesize[0];
744 }
745
746 /* draw the gauge using the momentary loudness */
747 p = pic->data[0] + ebur128->gauge.y*pic->linesize[0] + ebur128->gauge.x*3;
748 for (y = 0; y < ebur128->gauge.h; y++) {
749 const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_gauge, y);
750
751 for (x = 0; x < ebur128->gauge.w; x++)
752 memcpy(p + x*3, c, 3);
753 p += pic->linesize[0];
754 }
755
756 /* draw textual info */
757 drawtext(pic, PAD, PAD - PAD/2, FONT16, font_colors,
758 LOG_FMT " ", // padding to erase trailing characters
759 loudness_400, loudness_3000,
760 ebur128->integrated_loudness, ebur128->loudness_range);
761
762 /* set pts and push frame */
763 pic->pts = pts;
764 ret = ff_filter_frame(outlink, av_frame_clone(pic));
765 if (ret < 0)
766 return ret;
767 }
768
769 if (ebur128->metadata) { /* happens only once per filter_frame call */
770 char metabuf[128];
771#define META_PREFIX "lavfi.r128."
772
773#define SET_META(name, var) do { \
774 snprintf(metabuf, sizeof(metabuf), "%.3f", var); \
775 av_dict_set(&insamples->metadata, name, metabuf, 0); \
776} while (0)
777
778#define SET_META_PEAK(name, ptype) do { \
779 if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
780 char key[64]; \
781 for (ch = 0; ch < nb_channels; ch++) { \
782 snprintf(key, sizeof(key), \
783 META_PREFIX AV_STRINGIFY(name) "_peaks_ch%d", ch); \
784 SET_META(key, ebur128->name##_peaks[ch]); \
785 } \
786 } \
787} while (0)
788
789 SET_META(META_PREFIX "M", loudness_400);
790 SET_META(META_PREFIX "S", loudness_3000);
791 SET_META(META_PREFIX "I", ebur128->integrated_loudness);
792 SET_META(META_PREFIX "LRA", ebur128->loudness_range);
793 SET_META(META_PREFIX "LRA.low", ebur128->lra_low);
794 SET_META(META_PREFIX "LRA.high", ebur128->lra_high);
795
796 SET_META_PEAK(sample, SAMPLES);
797 SET_META_PEAK(true, TRUE);
798 }
799
800 av_log(ctx, ebur128->loglevel, "t: %-10s " LOG_FMT,
801 av_ts2timestr(pts, &outlink->time_base),
802 loudness_400, loudness_3000,
803 ebur128->integrated_loudness, ebur128->loudness_range);
804
805#define PRINT_PEAKS(str, sp, ptype) do { \
806 if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
807 av_log(ctx, ebur128->loglevel, " " str ":"); \
808 for (ch = 0; ch < nb_channels; ch++) \
809 av_log(ctx, ebur128->loglevel, " %5.1f", DBFS(sp[ch])); \
810 av_log(ctx, ebur128->loglevel, " dBFS"); \
811 } \
812} while (0)
813
814 PRINT_PEAKS("SPK", ebur128->sample_peaks, SAMPLES);
815 PRINT_PEAKS("FTPK", ebur128->true_peaks_per_frame, TRUE);
816 PRINT_PEAKS("TPK", ebur128->true_peaks, TRUE);
817 av_log(ctx, ebur128->loglevel, "\n");
818 }
819 }
820
821 return ff_filter_frame(ctx->outputs[ebur128->do_video], insamples);
822}
823
824static int query_formats(AVFilterContext *ctx)
825{
826 EBUR128Context *ebur128 = ctx->priv;
827 AVFilterFormats *formats;
828 AVFilterChannelLayouts *layouts;
829 AVFilterLink *inlink = ctx->inputs[0];
830 AVFilterLink *outlink = ctx->outputs[0];
831 int ret;
832
833 static const enum AVSampleFormat sample_fmts[] = { AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_NONE };
834 static const int input_srate[] = {48000, -1}; // ITU-R BS.1770 provides coeff only for 48kHz
835 static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_RGB24, AV_PIX_FMT_NONE };
836
837 /* set optional output video format */
838 if (ebur128->do_video) {
839 formats = ff_make_format_list(pix_fmts);
840 if ((ret = ff_formats_ref(formats, &outlink->in_formats)) < 0)
841 return ret;
842 outlink = ctx->outputs[1];
843 }
844
845 /* set input and output audio formats
846 * Note: ff_set_common_* functions are not used because they affect all the
847 * links, and thus break the video format negotiation */
848 formats = ff_make_format_list(sample_fmts);
849 if ((ret = ff_formats_ref(formats, &inlink->out_formats)) < 0 ||
850 (ret = ff_formats_ref(formats, &outlink->in_formats)) < 0)
851 return ret;
852
853 layouts = ff_all_channel_layouts();
854 if ((ret = ff_channel_layouts_ref(layouts, &inlink->out_channel_layouts)) < 0 ||
855 (ret = ff_channel_layouts_ref(layouts, &outlink->in_channel_layouts)) < 0)
856 return ret;
857
858 formats = ff_make_format_list(input_srate);
859 if ((ret = ff_formats_ref(formats, &inlink->out_samplerates)) < 0 ||
860 (ret = ff_formats_ref(formats, &outlink->in_samplerates)) < 0)
861 return ret;
862
863 return 0;
864}
865
866static av_cold void uninit(AVFilterContext *ctx)
867{
868 int i;
869 EBUR128Context *ebur128 = ctx->priv;
870
871 /* dual-mono correction */
872 if (ebur128->nb_channels == 1 && ebur128->dual_mono) {
873 ebur128->i400.rel_threshold -= ebur128->pan_law;
874 ebur128->i3000.rel_threshold -= ebur128->pan_law;
875 ebur128->lra_low -= ebur128->pan_law;
876 ebur128->lra_high -= ebur128->pan_law;
877 }
878
879 av_log(ctx, AV_LOG_INFO, "Summary:\n\n"
880 " Integrated loudness:\n"
881 " I: %5.1f LUFS\n"
882 " Threshold: %5.1f LUFS\n\n"
883 " Loudness range:\n"
884 " LRA: %5.1f LU\n"
885 " Threshold: %5.1f LUFS\n"
886 " LRA low: %5.1f LUFS\n"
887 " LRA high: %5.1f LUFS",
888 ebur128->integrated_loudness, ebur128->i400.rel_threshold,
889 ebur128->loudness_range, ebur128->i3000.rel_threshold,
890 ebur128->lra_low, ebur128->lra_high);
891
892#define PRINT_PEAK_SUMMARY(str, sp, ptype) do { \
893 int ch; \
894 double maxpeak; \
895 maxpeak = 0.0; \
896 if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
897 for (ch = 0; ch < ebur128->nb_channels; ch++) \
898 maxpeak = FFMAX(maxpeak, sp[ch]); \
899 av_log(ctx, AV_LOG_INFO, "\n\n " str " peak:\n" \
900 " Peak: %5.1f dBFS", \
901 DBFS(maxpeak)); \
902 } \
903} while (0)
904
905 PRINT_PEAK_SUMMARY("Sample", ebur128->sample_peaks, SAMPLES);
906 PRINT_PEAK_SUMMARY("True", ebur128->true_peaks, TRUE);
907 av_log(ctx, AV_LOG_INFO, "\n");
908
909 av_freep(&ebur128->y_line_ref);
910 av_freep(&ebur128->ch_weighting);
911 av_freep(&ebur128->true_peaks);
912 av_freep(&ebur128->sample_peaks);
913 av_freep(&ebur128->true_peaks_per_frame);
914 av_freep(&ebur128->i400.histogram);
915 av_freep(&ebur128->i3000.histogram);
916 for (i = 0; i < ebur128->nb_channels; i++) {
917 av_freep(&ebur128->i400.cache[i]);
918 av_freep(&ebur128->i3000.cache[i]);
919 }
920 for (i = 0; i < ctx->nb_outputs; i++)
921 av_freep(&ctx->output_pads[i].name);
922 av_frame_free(&ebur128->outpicref);
923#if CONFIG_SWRESAMPLE
924 av_freep(&ebur128->swr_buf);
925 swr_free(&ebur128->swr_ctx);
926#endif
927}
928
929static const AVFilterPad ebur128_inputs[] = {
930 {
931 .name = "default",
932 .type = AVMEDIA_TYPE_AUDIO,
933 .filter_frame = filter_frame,
934 .config_props = config_audio_input,
935 },
936 { NULL }
937};
938
939AVFilter ff_af_ebur128 = {
940 .name = "ebur128",
941 .description = NULL_IF_CONFIG_SMALL("EBU R128 scanner."),
942 .priv_size = sizeof(EBUR128Context),
943 .init = init,
944 .uninit = uninit,
945 .query_formats = query_formats,
946 .inputs = ebur128_inputs,
947 .outputs = NULL,
948 .priv_class = &ebur128_class,
949 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
950};
951
generated by cgit at 2024-05-17 13:01:23 (GMT)