2 * Copyright (c) 2012 Clément Bœsch
4 * This file is part of FFmpeg.
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.
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.
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
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
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"
46 #define MAX_CHANNELS 63
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
55 /* RLB-filter coefficients */
59 #define RLB_A1 -1.99004745483398
60 #define RLB_A2 0.99007225036621
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)
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.
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)
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
91 struct rect { int x, y, w, h; };
93 typedef struct EBUR128Context {
94 const AVClass *class; ///< AVClass context for log and options purpose
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
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_opt_max; ///< the y value (pixel position) for 1 LU
118 int y_opt_min; ///< the y value (pixel position) for -1 LU
119 int *y_line_ref; ///< y reference values for drawing the LU lines in the graph and the gauge
122 int nb_channels; ///< number of channels in the input
123 double *ch_weighting; ///< channel weighting mapping
124 int sample_count; ///< sample count used for refresh frequency, reset at refresh
127 * The mult by 3 in the following is for X[i], X[i-1] and X[i-2] */
128 double x[MAX_CHANNELS * 3]; ///< 3 input samples cache for each channel
129 double y[MAX_CHANNELS * 3]; ///< 3 pre-filter samples cache for each channel
130 double z[MAX_CHANNELS * 3]; ///< 3 RLB-filter samples cache for each channel
132 #define I400_BINS (48000 * 4 / 10)
133 #define I3000_BINS (48000 * 3)
134 struct integrator i400; ///< 400ms integrator, used for Momentary loudness (M), and Integrated loudness (I)
135 struct integrator i3000; ///< 3s integrator, used for Short term loudness (S), and Loudness Range (LRA)
137 /* I and LRA specific */
138 double integrated_loudness; ///< integrated loudness in LUFS (I)
139 double loudness_range; ///< loudness range in LU (LRA)
140 double lra_low, lra_high; ///< low and high LRA values
143 int loglevel; ///< log level for frame logging
144 int metadata; ///< whether or not to inject loudness results in frames
145 int dual_mono; ///< whether or not to treat single channel input files as dual-mono
146 double pan_law; ///< pan law value used to calculate dual-mono measurements
147 int target; ///< target level in LUFS used to set relative zero LU in visualization
148 int gauge_type; ///< whether gauge shows momentary or short
153 PEAK_MODE_SAMPLES_PEAKS = 1<<1,
154 PEAK_MODE_TRUE_PEAKS = 1<<2,
158 GAUGE_TYPE_MOMENTARY = 0,
159 GAUGE_TYPE_SHORTTERM = 1,
163 #define OFFSET(x) offsetof(EBUR128Context, x)
164 #define A AV_OPT_FLAG_AUDIO_PARAM
165 #define V AV_OPT_FLAG_VIDEO_PARAM
166 #define F AV_OPT_FLAG_FILTERING_PARAM
167 static const AVOption ebur128_options[] = {
168 { "video", "set video output", OFFSET(do_video), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, V|F },
169 { "size", "set video size", OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "640x480"}, 0, 0, V|F },
170 { "meter", "set scale meter (+9 to +18)", OFFSET(meter), AV_OPT_TYPE_INT, {.i64 = 9}, 9, 18, V|F },
171 { "framelog", "force frame logging level", OFFSET(loglevel), AV_OPT_TYPE_INT, {.i64 = -1}, INT_MIN, INT_MAX, A|V|F, "level" },
172 { "info", "information logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_INFO}, INT_MIN, INT_MAX, A|V|F, "level" },
173 { "verbose", "verbose logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_VERBOSE}, INT_MIN, INT_MAX, A|V|F, "level" },
174 { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, A|V|F },
175 { "peak", "set peak mode", OFFSET(peak_mode), AV_OPT_TYPE_FLAGS, {.i64 = PEAK_MODE_NONE}, 0, INT_MAX, A|F, "mode" },
176 { "none", "disable any peak mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_NONE}, INT_MIN, INT_MAX, A|F, "mode" },
177 { "sample", "enable peak-sample mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_SAMPLES_PEAKS}, INT_MIN, INT_MAX, A|F, "mode" },
178 { "true", "enable true-peak mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_TRUE_PEAKS}, INT_MIN, INT_MAX, A|F, "mode" },
179 { "dualmono", "treat mono input files as dual-mono", OFFSET(dual_mono), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, A|F },
180 { "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 },
181 { "target", "set a specific target level in LUFS (-23 to 0)", OFFSET(target), AV_OPT_TYPE_INT, {.i64 = -23}, -23, 0, V|F },
182 { "gauge", "set gauge display type", OFFSET(gauge_type), AV_OPT_TYPE_INT, {.i64 = 0 }, GAUGE_TYPE_MOMENTARY, GAUGE_TYPE_SHORTTERM, V|F, "gaugetype" },
183 { "momentary", "display momentary value", 0, AV_OPT_TYPE_CONST, {.i64 = GAUGE_TYPE_MOMENTARY}, INT_MIN, INT_MAX, V|F, "gaugetype" },
184 { "m", "display momentary value", 0, AV_OPT_TYPE_CONST, {.i64 = GAUGE_TYPE_MOMENTARY}, INT_MIN, INT_MAX, V|F, "gaugetype" },
185 { "shortterm", "display short-term value", 0, AV_OPT_TYPE_CONST, {.i64 = GAUGE_TYPE_SHORTTERM}, INT_MIN, INT_MAX, V|F, "gaugetype" },
186 { "s", "display short-term value", 0, AV_OPT_TYPE_CONST, {.i64 = GAUGE_TYPE_SHORTTERM}, INT_MIN, INT_MAX, V|F, "gaugetype" },
190 AVFILTER_DEFINE_CLASS(ebur128);
192 static const uint8_t graph_colors[] = {
193 0xdd, 0x66, 0x66, // value above 1LU non reached below -1LU (impossible)
194 0x66, 0x66, 0xdd, // value below 1LU non reached below -1LU
195 0x96, 0x33, 0x33, // value above 1LU reached below -1LU (impossible)
196 0x33, 0x33, 0x96, // value below 1LU reached below -1LU
197 0xdd, 0x96, 0x96, // value above 1LU line non reached below -1LU (impossible)
198 0x96, 0x96, 0xdd, // value below 1LU line non reached below -1LU
199 0xdd, 0x33, 0x33, // value above 1LU line reached below -1LU (impossible)
200 0x33, 0x33, 0xdd, // value below 1LU line reached below -1LU
201 0xdd, 0x66, 0x66, // value above 1LU non reached above -1LU
202 0x66, 0xdd, 0x66, // value below 1LU non reached above -1LU
203 0x96, 0x33, 0x33, // value above 1LU reached above -1LU
204 0x33, 0x96, 0x33, // value below 1LU reached above -1LU
205 0xdd, 0x96, 0x96, // value above 1LU line non reached above -1LU
206 0x96, 0xdd, 0x96, // value below 1LU line non reached above -1LU
207 0xdd, 0x33, 0x33, // value above 1LU line reached above -1LU
208 0x33, 0xdd, 0x33, // value below 1LU line reached above -1LU
211 static const uint8_t *get_graph_color(const EBUR128Context *ebur128, int v, int y)
213 const int above_opt_max = y > ebur128->y_opt_max;
214 const int below_opt_min = y < ebur128->y_opt_min;
215 const int reached = y >= v;
216 const int line = ebur128->y_line_ref[y] || y == ebur128->y_zero_lu;
217 const int colorid = 8*below_opt_min+ 4*line + 2*reached + above_opt_max;
218 return graph_colors + 3*colorid;
221 static inline int lu_to_y(const EBUR128Context *ebur128, double v)
223 v += 2 * ebur128->meter; // make it in range [0;...]
224 v = av_clipf(v, 0, ebur128->scale_range); // make sure it's in the graph scale
225 v = ebur128->scale_range - v; // invert value (y=0 is on top)
226 return v * ebur128->graph.h / ebur128->scale_range; // rescale from scale range to px height
232 static const uint8_t font_colors[] = {
237 static void drawtext(AVFrame *pic, int x, int y, int ftid, const uint8_t *color, const char *fmt, ...)
245 if (ftid == FONT16) font = avpriv_vga16_font, font_height = 16;
246 else if (ftid == FONT8) font = avpriv_cga_font, font_height = 8;
250 vsnprintf(buf, sizeof(buf), fmt, vl);
253 for (i = 0; buf[i]; i++) {
255 uint8_t *p = pic->data[0] + y*pic->linesize[0] + (x + i*8)*3;
257 for (char_y = 0; char_y < font_height; char_y++) {
258 for (mask = 0x80; mask; mask >>= 1) {
259 if (font[buf[i] * font_height + char_y] & mask)
262 memcpy(p, "\x00\x00\x00", 3);
265 p += pic->linesize[0] - 8*3;
270 static void drawline(AVFrame *pic, int x, int y, int len, int step)
273 uint8_t *p = pic->data[0] + y*pic->linesize[0] + x*3;
275 for (i = 0; i < len; i++) {
276 memcpy(p, "\x00\xff\x00", 3);
281 static int config_video_output(AVFilterLink *outlink)
285 AVFilterContext *ctx = outlink->src;
286 EBUR128Context *ebur128 = ctx->priv;
289 /* check if there is enough space to represent everything decently */
290 if (ebur128->w < 640 || ebur128->h < 480) {
291 av_log(ctx, AV_LOG_ERROR, "Video size %dx%d is too small, "
292 "minimum size is 640x480\n", ebur128->w, ebur128->h);
293 return AVERROR(EINVAL);
295 outlink->w = ebur128->w;
296 outlink->h = ebur128->h;
297 outlink->sample_aspect_ratio = (AVRational){1,1};
301 /* configure text area position and size */
302 ebur128->text.x = PAD;
303 ebur128->text.y = 40;
304 ebur128->text.w = 3 * 8; // 3 characters
305 ebur128->text.h = ebur128->h - PAD - ebur128->text.y;
307 /* configure gauge position and size */
308 ebur128->gauge.w = 20;
309 ebur128->gauge.h = ebur128->text.h;
310 ebur128->gauge.x = ebur128->w - PAD - ebur128->gauge.w;
311 ebur128->gauge.y = ebur128->text.y;
313 /* configure graph position and size */
314 ebur128->graph.x = ebur128->text.x + ebur128->text.w + PAD;
315 ebur128->graph.y = ebur128->gauge.y;
316 ebur128->graph.w = ebur128->gauge.x - ebur128->graph.x - PAD;
317 ebur128->graph.h = ebur128->gauge.h;
319 /* graph and gauge share the LU-to-pixel code */
320 av_assert0(ebur128->graph.h == ebur128->gauge.h);
322 /* prepare the initial picref buffer */
323 av_frame_free(&ebur128->outpicref);
324 ebur128->outpicref = outpicref =
325 ff_get_video_buffer(outlink, outlink->w, outlink->h);
327 return AVERROR(ENOMEM);
328 outpicref->sample_aspect_ratio = (AVRational){1,1};
330 /* init y references values (to draw LU lines) */
331 ebur128->y_line_ref = av_calloc(ebur128->graph.h + 1, sizeof(*ebur128->y_line_ref));
332 if (!ebur128->y_line_ref)
333 return AVERROR(ENOMEM);
335 /* black background */
336 memset(outpicref->data[0], 0, ebur128->h * outpicref->linesize[0]);
338 /* draw LU legends */
339 drawtext(outpicref, PAD, PAD+16, FONT8, font_colors+3, " LU");
340 for (i = ebur128->meter; i >= -ebur128->meter * 2; i--) {
341 y = lu_to_y(ebur128, i);
342 x = PAD + (i < 10 && i > -10) * 8;
343 ebur128->y_line_ref[y] = i;
344 y -= 4; // -4 to center vertically
345 drawtext(outpicref, x, y + ebur128->graph.y, FONT8, font_colors+3,
346 "%c%d", i < 0 ? '-' : i > 0 ? '+' : ' ', FFABS(i));
350 ebur128->y_zero_lu = lu_to_y(ebur128, 0);
351 ebur128->y_opt_max = lu_to_y(ebur128, 1);
352 ebur128->y_opt_min = lu_to_y(ebur128, -1);
353 p = outpicref->data[0] + ebur128->graph.y * outpicref->linesize[0]
354 + ebur128->graph.x * 3;
355 for (y = 0; y < ebur128->graph.h; y++) {
356 const uint8_t *c = get_graph_color(ebur128, INT_MAX, y);
358 for (x = 0; x < ebur128->graph.w; x++)
359 memcpy(p + x*3, c, 3);
360 p += outpicref->linesize[0];
363 /* draw fancy rectangles around the graph and the gauge */
364 #define DRAW_RECT(r) do { \
365 drawline(outpicref, r.x, r.y - 1, r.w, 3); \
366 drawline(outpicref, r.x, r.y + r.h, r.w, 3); \
367 drawline(outpicref, r.x - 1, r.y, r.h, outpicref->linesize[0]); \
368 drawline(outpicref, r.x + r.w, r.y, r.h, outpicref->linesize[0]); \
370 DRAW_RECT(ebur128->graph);
371 DRAW_RECT(ebur128->gauge);
376 static int config_audio_input(AVFilterLink *inlink)
378 AVFilterContext *ctx = inlink->dst;
379 EBUR128Context *ebur128 = ctx->priv;
381 /* Force 100ms framing in case of metadata injection: the frames must have
382 * a granularity of the window overlap to be accurately exploited.
383 * As for the true peaks mode, it just simplifies the resampling buffer
384 * allocation and the lookup in it (since sample buffers differ in size, it
385 * can be more complex to integrate in the one-sample loop of
386 * filter_frame()). */
387 if (ebur128->metadata || (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS))
388 inlink->min_samples =
389 inlink->max_samples =
390 inlink->partial_buf_size = inlink->sample_rate / 10;
394 static int config_audio_output(AVFilterLink *outlink)
397 AVFilterContext *ctx = outlink->src;
398 EBUR128Context *ebur128 = ctx->priv;
399 const int nb_channels = av_get_channel_layout_nb_channels(outlink->channel_layout);
401 #define BACK_MASK (AV_CH_BACK_LEFT |AV_CH_BACK_CENTER |AV_CH_BACK_RIGHT| \
402 AV_CH_TOP_BACK_LEFT|AV_CH_TOP_BACK_CENTER|AV_CH_TOP_BACK_RIGHT| \
403 AV_CH_SIDE_LEFT |AV_CH_SIDE_RIGHT| \
404 AV_CH_SURROUND_DIRECT_LEFT |AV_CH_SURROUND_DIRECT_RIGHT)
406 ebur128->nb_channels = nb_channels;
407 ebur128->ch_weighting = av_calloc(nb_channels, sizeof(*ebur128->ch_weighting));
408 if (!ebur128->ch_weighting)
409 return AVERROR(ENOMEM);
411 for (i = 0; i < nb_channels; i++) {
412 /* channel weighting */
413 const uint16_t chl = av_channel_layout_extract_channel(outlink->channel_layout, i);
414 if (chl & (AV_CH_LOW_FREQUENCY|AV_CH_LOW_FREQUENCY_2)) {
415 ebur128->ch_weighting[i] = 0;
416 } else if (chl & BACK_MASK) {
417 ebur128->ch_weighting[i] = 1.41;
419 ebur128->ch_weighting[i] = 1.0;
422 if (!ebur128->ch_weighting[i])
425 /* bins buffer for the two integration window (400ms and 3s) */
426 ebur128->i400.cache[i] = av_calloc(I400_BINS, sizeof(*ebur128->i400.cache[0]));
427 ebur128->i3000.cache[i] = av_calloc(I3000_BINS, sizeof(*ebur128->i3000.cache[0]));
428 if (!ebur128->i400.cache[i] || !ebur128->i3000.cache[i])
429 return AVERROR(ENOMEM);
432 #if CONFIG_SWRESAMPLE
433 if (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS) {
436 ebur128->swr_buf = av_malloc_array(nb_channels, 19200 * sizeof(double));
437 ebur128->true_peaks = av_calloc(nb_channels, sizeof(*ebur128->true_peaks));
438 ebur128->true_peaks_per_frame = av_calloc(nb_channels, sizeof(*ebur128->true_peaks_per_frame));
439 ebur128->swr_ctx = swr_alloc();
440 if (!ebur128->swr_buf || !ebur128->true_peaks ||
441 !ebur128->true_peaks_per_frame || !ebur128->swr_ctx)
442 return AVERROR(ENOMEM);
444 av_opt_set_int(ebur128->swr_ctx, "in_channel_layout", outlink->channel_layout, 0);
445 av_opt_set_int(ebur128->swr_ctx, "in_sample_rate", outlink->sample_rate, 0);
446 av_opt_set_sample_fmt(ebur128->swr_ctx, "in_sample_fmt", outlink->format, 0);
448 av_opt_set_int(ebur128->swr_ctx, "out_channel_layout", outlink->channel_layout, 0);
449 av_opt_set_int(ebur128->swr_ctx, "out_sample_rate", 192000, 0);
450 av_opt_set_sample_fmt(ebur128->swr_ctx, "out_sample_fmt", outlink->format, 0);
452 ret = swr_init(ebur128->swr_ctx);
458 if (ebur128->peak_mode & PEAK_MODE_SAMPLES_PEAKS) {
459 ebur128->sample_peaks = av_calloc(nb_channels, sizeof(*ebur128->sample_peaks));
460 if (!ebur128->sample_peaks)
461 return AVERROR(ENOMEM);
467 #define ENERGY(loudness) (ff_exp10(((loudness) + 0.691) / 10.))
468 #define LOUDNESS(energy) (-0.691 + 10 * log10(energy))
469 #define DBFS(energy) (20 * log10(energy))
471 static struct hist_entry *get_histogram(void)
474 struct hist_entry *h = av_calloc(HIST_SIZE, sizeof(*h));
478 for (i = 0; i < HIST_SIZE; i++) {
479 h[i].loudness = i / (double)HIST_GRAIN + ABS_THRES;
480 h[i].energy = ENERGY(h[i].loudness);
485 static av_cold int init(AVFilterContext *ctx)
487 EBUR128Context *ebur128 = ctx->priv;
491 if (ebur128->loglevel != AV_LOG_INFO &&
492 ebur128->loglevel != AV_LOG_VERBOSE) {
493 if (ebur128->do_video || ebur128->metadata)
494 ebur128->loglevel = AV_LOG_VERBOSE;
496 ebur128->loglevel = AV_LOG_INFO;
499 if (!CONFIG_SWRESAMPLE && (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS)) {
500 av_log(ctx, AV_LOG_ERROR,
501 "True-peak mode requires libswresample to be performed\n");
502 return AVERROR(EINVAL);
505 // if meter is +9 scale, scale range is from -18 LU to +9 LU (or 3*9)
506 // if meter is +18 scale, scale range is from -36 LU to +18 LU (or 3*18)
507 ebur128->scale_range = 3 * ebur128->meter;
509 ebur128->i400.histogram = get_histogram();
510 ebur128->i3000.histogram = get_histogram();
511 if (!ebur128->i400.histogram || !ebur128->i3000.histogram)
512 return AVERROR(ENOMEM);
514 ebur128->integrated_loudness = ABS_THRES;
515 ebur128->loudness_range = 0;
517 /* insert output pads */
518 if (ebur128->do_video) {
520 .name = av_strdup("out0"),
521 .type = AVMEDIA_TYPE_VIDEO,
522 .config_props = config_video_output,
525 return AVERROR(ENOMEM);
526 ret = ff_insert_outpad(ctx, 0, &pad);
533 .name = av_asprintf("out%d", ebur128->do_video),
534 .type = AVMEDIA_TYPE_AUDIO,
535 .config_props = config_audio_output,
538 return AVERROR(ENOMEM);
539 ret = ff_insert_outpad(ctx, ebur128->do_video, &pad);
546 av_log(ctx, AV_LOG_VERBOSE, "EBU +%d scale\n", ebur128->meter);
551 #define HIST_POS(power) (int)(((power) - ABS_THRES) * HIST_GRAIN)
553 /* loudness and power should be set such as loudness = -0.691 +
554 * 10*log10(power), we just avoid doing that calculus two times */
555 static int gate_update(struct integrator *integ, double power,
556 double loudness, int gate_thres)
559 double relative_threshold;
562 /* update powers histograms by incrementing current power count */
563 ipower = av_clip(HIST_POS(loudness), 0, HIST_SIZE - 1);
564 integ->histogram[ipower].count++;
566 /* compute relative threshold and get its position in the histogram */
567 integ->sum_kept_powers += power;
568 integ->nb_kept_powers++;
569 relative_threshold = integ->sum_kept_powers / integ->nb_kept_powers;
570 if (!relative_threshold)
571 relative_threshold = 1e-12;
572 integ->rel_threshold = LOUDNESS(relative_threshold) + gate_thres;
573 gate_hist_pos = av_clip(HIST_POS(integ->rel_threshold), 0, HIST_SIZE - 1);
575 return gate_hist_pos;
578 static int filter_frame(AVFilterLink *inlink, AVFrame *insamples)
580 int i, ch, idx_insample;
581 AVFilterContext *ctx = inlink->dst;
582 EBUR128Context *ebur128 = ctx->priv;
583 const int nb_channels = ebur128->nb_channels;
584 const int nb_samples = insamples->nb_samples;
585 const double *samples = (double *)insamples->data[0];
586 AVFrame *pic = ebur128->outpicref;
588 #if CONFIG_SWRESAMPLE
589 if (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS) {
590 const double *swr_samples = ebur128->swr_buf;
591 int ret = swr_convert(ebur128->swr_ctx, (uint8_t**)&ebur128->swr_buf, 19200,
592 (const uint8_t **)insamples->data, nb_samples);
595 for (ch = 0; ch < nb_channels; ch++)
596 ebur128->true_peaks_per_frame[ch] = 0.0;
597 for (idx_insample = 0; idx_insample < ret; idx_insample++) {
598 for (ch = 0; ch < nb_channels; ch++) {
599 ebur128->true_peaks[ch] = FFMAX(ebur128->true_peaks[ch], fabs(*swr_samples));
600 ebur128->true_peaks_per_frame[ch] = FFMAX(ebur128->true_peaks_per_frame[ch],
608 for (idx_insample = 0; idx_insample < nb_samples; idx_insample++) {
609 const int bin_id_400 = ebur128->i400.cache_pos;
610 const int bin_id_3000 = ebur128->i3000.cache_pos;
612 #define MOVE_TO_NEXT_CACHED_ENTRY(time) do { \
613 ebur128->i##time.cache_pos++; \
614 if (ebur128->i##time.cache_pos == I##time##_BINS) { \
615 ebur128->i##time.filled = 1; \
616 ebur128->i##time.cache_pos = 0; \
620 MOVE_TO_NEXT_CACHED_ENTRY(400);
621 MOVE_TO_NEXT_CACHED_ENTRY(3000);
623 for (ch = 0; ch < nb_channels; ch++) {
626 if (ebur128->peak_mode & PEAK_MODE_SAMPLES_PEAKS)
627 ebur128->sample_peaks[ch] = FFMAX(ebur128->sample_peaks[ch], fabs(*samples));
629 ebur128->x[ch * 3] = *samples++; // set X[i]
631 if (!ebur128->ch_weighting[ch])
634 /* Y[i] = X[i]*b0 + X[i-1]*b1 + X[i-2]*b2 - Y[i-1]*a1 - Y[i-2]*a2 */
635 #define FILTER(Y, X, name) do { \
636 double *dst = ebur128->Y + ch*3; \
637 double *src = ebur128->X + ch*3; \
640 dst[0] = src[0]*name##_B0 + src[1]*name##_B1 + src[2]*name##_B2 \
641 - dst[1]*name##_A1 - dst[2]*name##_A2; \
644 // TODO: merge both filters in one?
645 FILTER(y, x, PRE); // apply pre-filter
646 ebur128->x[ch * 3 + 2] = ebur128->x[ch * 3 + 1];
647 ebur128->x[ch * 3 + 1] = ebur128->x[ch * 3 ];
648 FILTER(z, y, RLB); // apply RLB-filter
650 bin = ebur128->z[ch * 3] * ebur128->z[ch * 3];
652 /* add the new value, and limit the sum to the cache size (400ms or 3s)
653 * by removing the oldest one */
654 ebur128->i400.sum [ch] = ebur128->i400.sum [ch] + bin - ebur128->i400.cache [ch][bin_id_400];
655 ebur128->i3000.sum[ch] = ebur128->i3000.sum[ch] + bin - ebur128->i3000.cache[ch][bin_id_3000];
657 /* override old cache entry with the new value */
658 ebur128->i400.cache [ch][bin_id_400 ] = bin;
659 ebur128->i3000.cache[ch][bin_id_3000] = bin;
662 /* For integrated loudness, gating blocks are 400ms long with 75%
663 * overlap (see BS.1770-2 p5), so a re-computation is needed each 100ms
664 * (4800 samples at 48kHz). */
665 if (++ebur128->sample_count == 4800) {
666 double loudness_400, loudness_3000;
667 double power_400 = 1e-12, power_3000 = 1e-12;
668 AVFilterLink *outlink = ctx->outputs[0];
669 const int64_t pts = insamples->pts +
670 av_rescale_q(idx_insample, (AVRational){ 1, inlink->sample_rate },
673 ebur128->sample_count = 0;
675 #define COMPUTE_LOUDNESS(m, time) do { \
676 if (ebur128->i##time.filled) { \
677 /* weighting sum of the last <time> ms */ \
678 for (ch = 0; ch < nb_channels; ch++) \
679 power_##time += ebur128->ch_weighting[ch] * ebur128->i##time.sum[ch]; \
680 power_##time /= I##time##_BINS; \
682 loudness_##time = LOUDNESS(power_##time); \
685 COMPUTE_LOUDNESS(M, 400);
686 COMPUTE_LOUDNESS(S, 3000);
688 /* Integrated loudness */
689 #define I_GATE_THRES -10 // initially defined to -8 LU in the first EBU standard
691 if (loudness_400 >= ABS_THRES) {
692 double integrated_sum = 0;
693 int nb_integrated = 0;
694 int gate_hist_pos = gate_update(&ebur128->i400, power_400,
695 loudness_400, I_GATE_THRES);
697 /* compute integrated loudness by summing the histogram values
698 * above the relative threshold */
699 for (i = gate_hist_pos; i < HIST_SIZE; i++) {
700 const int nb_v = ebur128->i400.histogram[i].count;
701 nb_integrated += nb_v;
702 integrated_sum += nb_v * ebur128->i400.histogram[i].energy;
705 ebur128->integrated_loudness = LOUDNESS(integrated_sum / nb_integrated);
706 /* dual-mono correction */
707 if (nb_channels == 1 && ebur128->dual_mono) {
708 ebur128->integrated_loudness -= ebur128->pan_law;
714 #define LRA_GATE_THRES -20
715 #define LRA_LOWER_PRC 10
716 #define LRA_HIGHER_PRC 95
718 /* XXX: example code in EBU 3342 is ">=" but formula in BS.1770
720 if (loudness_3000 >= ABS_THRES) {
722 int gate_hist_pos = gate_update(&ebur128->i3000, power_3000,
723 loudness_3000, LRA_GATE_THRES);
725 for (i = gate_hist_pos; i < HIST_SIZE; i++)
726 nb_powers += ebur128->i3000.histogram[i].count;
730 /* get lower loudness to consider */
732 nb_pow = LRA_LOWER_PRC * nb_powers / 100. + 0.5;
733 for (i = gate_hist_pos; i < HIST_SIZE; i++) {
734 n += ebur128->i3000.histogram[i].count;
736 ebur128->lra_low = ebur128->i3000.histogram[i].loudness;
741 /* get higher loudness to consider */
743 nb_pow = LRA_HIGHER_PRC * nb_powers / 100. + 0.5;
744 for (i = HIST_SIZE - 1; i >= 0; i--) {
745 n -= ebur128->i3000.histogram[i].count;
747 ebur128->lra_high = ebur128->i3000.histogram[i].loudness;
752 // XXX: show low & high on the graph?
753 ebur128->loudness_range = ebur128->lra_high - ebur128->lra_low;
757 /* dual-mono correction */
758 if (nb_channels == 1 && ebur128->dual_mono) {
759 loudness_400 -= ebur128->pan_law;
760 loudness_3000 -= ebur128->pan_law;
763 #define LOG_FMT "TARGET:%d M:%6.1f S:%6.1f I:%6.1f LUFS LRA:%6.1f LU"
765 /* push one video frame */
766 if (ebur128->do_video) {
771 if (ebur128->gauge_type == GAUGE_TYPE_MOMENTARY) {
772 gauge_value = loudness_400 - ebur128->target;
774 gauge_value = loudness_3000 - ebur128->target;
777 const int y_loudness_lu_graph = lu_to_y(ebur128, loudness_3000 - ebur128->target);
778 const int y_loudness_lu_gauge = lu_to_y(ebur128, gauge_value);
780 /* draw the graph using the short-term loudness */
781 p = pic->data[0] + ebur128->graph.y*pic->linesize[0] + ebur128->graph.x*3;
782 for (y = 0; y < ebur128->graph.h; y++) {
783 const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_graph, y);
785 memmove(p, p + 3, (ebur128->graph.w - 1) * 3);
786 memcpy(p + (ebur128->graph.w - 1) * 3, c, 3);
787 p += pic->linesize[0];
790 /* draw the gauge using either momentary or short-term loudness */
791 p = pic->data[0] + ebur128->gauge.y*pic->linesize[0] + ebur128->gauge.x*3;
792 for (y = 0; y < ebur128->gauge.h; y++) {
793 const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_gauge, y);
795 for (x = 0; x < ebur128->gauge.w; x++)
796 memcpy(p + x*3, c, 3);
797 p += pic->linesize[0];
800 /* draw textual info */
801 drawtext(pic, PAD, PAD - PAD/2, FONT16, font_colors,
802 LOG_FMT " ", // padding to erase trailing characters
803 ebur128->target, loudness_400, loudness_3000,
804 ebur128->integrated_loudness, ebur128->loudness_range);
806 /* set pts and push frame */
808 ret = ff_filter_frame(outlink, av_frame_clone(pic));
813 if (ebur128->metadata) { /* happens only once per filter_frame call */
815 #define META_PREFIX "lavfi.r128."
817 #define SET_META(name, var) do { \
818 snprintf(metabuf, sizeof(metabuf), "%.3f", var); \
819 av_dict_set(&insamples->metadata, name, metabuf, 0); \
822 #define SET_META_PEAK(name, ptype) do { \
823 if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
825 for (ch = 0; ch < nb_channels; ch++) { \
826 snprintf(key, sizeof(key), \
827 META_PREFIX AV_STRINGIFY(name) "_peaks_ch%d", ch); \
828 SET_META(key, ebur128->name##_peaks[ch]); \
833 SET_META(META_PREFIX "M", loudness_400);
834 SET_META(META_PREFIX "S", loudness_3000);
835 SET_META(META_PREFIX "I", ebur128->integrated_loudness);
836 SET_META(META_PREFIX "LRA", ebur128->loudness_range);
837 SET_META(META_PREFIX "LRA.low", ebur128->lra_low);
838 SET_META(META_PREFIX "LRA.high", ebur128->lra_high);
840 SET_META_PEAK(sample, SAMPLES);
841 SET_META_PEAK(true, TRUE);
844 av_log(ctx, ebur128->loglevel, "t: %-10s " LOG_FMT,
845 av_ts2timestr(pts, &outlink->time_base),
846 ebur128->target, loudness_400, loudness_3000,
847 ebur128->integrated_loudness, ebur128->loudness_range);
849 #define PRINT_PEAKS(str, sp, ptype) do { \
850 if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
851 av_log(ctx, ebur128->loglevel, " " str ":"); \
852 for (ch = 0; ch < nb_channels; ch++) \
853 av_log(ctx, ebur128->loglevel, " %5.1f", DBFS(sp[ch])); \
854 av_log(ctx, ebur128->loglevel, " dBFS"); \
858 PRINT_PEAKS("SPK", ebur128->sample_peaks, SAMPLES);
859 PRINT_PEAKS("FTPK", ebur128->true_peaks_per_frame, TRUE);
860 PRINT_PEAKS("TPK", ebur128->true_peaks, TRUE);
861 av_log(ctx, ebur128->loglevel, "\n");
865 return ff_filter_frame(ctx->outputs[ebur128->do_video], insamples);
868 static int query_formats(AVFilterContext *ctx)
870 EBUR128Context *ebur128 = ctx->priv;
871 AVFilterFormats *formats;
872 AVFilterChannelLayouts *layouts;
873 AVFilterLink *inlink = ctx->inputs[0];
874 AVFilterLink *outlink = ctx->outputs[0];
877 static const enum AVSampleFormat sample_fmts[] = { AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_NONE };
878 static const int input_srate[] = {48000, -1}; // ITU-R BS.1770 provides coeff only for 48kHz
879 static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_RGB24, AV_PIX_FMT_NONE };
881 /* set optional output video format */
882 if (ebur128->do_video) {
883 formats = ff_make_format_list(pix_fmts);
884 if ((ret = ff_formats_ref(formats, &outlink->in_formats)) < 0)
886 outlink = ctx->outputs[1];
889 /* set input and output audio formats
890 * Note: ff_set_common_* functions are not used because they affect all the
891 * links, and thus break the video format negotiation */
892 formats = ff_make_format_list(sample_fmts);
893 if ((ret = ff_formats_ref(formats, &inlink->out_formats)) < 0 ||
894 (ret = ff_formats_ref(formats, &outlink->in_formats)) < 0)
897 layouts = ff_all_channel_layouts();
898 if ((ret = ff_channel_layouts_ref(layouts, &inlink->out_channel_layouts)) < 0 ||
899 (ret = ff_channel_layouts_ref(layouts, &outlink->in_channel_layouts)) < 0)
902 formats = ff_make_format_list(input_srate);
903 if ((ret = ff_formats_ref(formats, &inlink->out_samplerates)) < 0 ||
904 (ret = ff_formats_ref(formats, &outlink->in_samplerates)) < 0)
910 static av_cold void uninit(AVFilterContext *ctx)
913 EBUR128Context *ebur128 = ctx->priv;
915 /* dual-mono correction */
916 if (ebur128->nb_channels == 1 && ebur128->dual_mono) {
917 ebur128->i400.rel_threshold -= ebur128->pan_law;
918 ebur128->i3000.rel_threshold -= ebur128->pan_law;
919 ebur128->lra_low -= ebur128->pan_law;
920 ebur128->lra_high -= ebur128->pan_law;
923 av_log(ctx, AV_LOG_INFO, "Summary:\n\n"
924 " Integrated loudness:\n"
926 " Threshold: %5.1f LUFS\n\n"
929 " Threshold: %5.1f LUFS\n"
930 " LRA low: %5.1f LUFS\n"
931 " LRA high: %5.1f LUFS",
932 ebur128->integrated_loudness, ebur128->i400.rel_threshold,
933 ebur128->loudness_range, ebur128->i3000.rel_threshold,
934 ebur128->lra_low, ebur128->lra_high);
936 #define PRINT_PEAK_SUMMARY(str, sp, ptype) do { \
940 if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
941 for (ch = 0; ch < ebur128->nb_channels; ch++) \
942 maxpeak = FFMAX(maxpeak, sp[ch]); \
943 av_log(ctx, AV_LOG_INFO, "\n\n " str " peak:\n" \
944 " Peak: %5.1f dBFS", \
949 PRINT_PEAK_SUMMARY("Sample", ebur128->sample_peaks, SAMPLES);
950 PRINT_PEAK_SUMMARY("True", ebur128->true_peaks, TRUE);
951 av_log(ctx, AV_LOG_INFO, "\n");
953 av_freep(&ebur128->y_line_ref);
954 av_freep(&ebur128->ch_weighting);
955 av_freep(&ebur128->true_peaks);
956 av_freep(&ebur128->sample_peaks);
957 av_freep(&ebur128->true_peaks_per_frame);
958 av_freep(&ebur128->i400.histogram);
959 av_freep(&ebur128->i3000.histogram);
960 for (i = 0; i < ebur128->nb_channels; i++) {
961 av_freep(&ebur128->i400.cache[i]);
962 av_freep(&ebur128->i3000.cache[i]);
964 for (i = 0; i < ctx->nb_outputs; i++)
965 av_freep(&ctx->output_pads[i].name);
966 av_frame_free(&ebur128->outpicref);
967 #if CONFIG_SWRESAMPLE
968 av_freep(&ebur128->swr_buf);
969 swr_free(&ebur128->swr_ctx);
973 static const AVFilterPad ebur128_inputs[] = {
976 .type = AVMEDIA_TYPE_AUDIO,
977 .filter_frame = filter_frame,
978 .config_props = config_audio_input,
983 AVFilter ff_af_ebur128 = {
985 .description = NULL_IF_CONFIG_SMALL("EBU R128 scanner."),
986 .priv_size = sizeof(EBUR128Context),
989 .query_formats = query_formats,
990 .inputs = ebur128_inputs,
992 .priv_class = &ebur128_class,
993 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,