yading@10: /*
yading@10:  * Copyright (c) 2012 Clément Bœsch
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or modify
yading@10:  * it under the terms of the GNU General Public License as published by
yading@10:  * the Free Software Foundation; either version 2 of the License, or
yading@10:  * (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
yading@10:  * GNU General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU General Public License along
yading@10:  * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
yading@10:  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
yading@10:  */
yading@10: 
yading@10: /**
yading@10:  * @file
yading@10:  * EBU R.128 implementation
yading@10:  * @see http://tech.ebu.ch/loudness
yading@10:  * @see https://www.youtube.com/watch?v=iuEtQqC-Sqo "EBU R128 Introduction - Florian Camerer"
yading@10:  * @todo True Peak
yading@10:  * @todo implement start/stop/reset through filter command injection
yading@10:  * @todo support other frequencies to avoid resampling
yading@10:  */
yading@10: 
yading@10: #include <math.h>
yading@10: 
yading@10: #include "libavutil/avassert.h"
yading@10: #include "libavutil/avstring.h"
yading@10: #include "libavutil/channel_layout.h"
yading@10: #include "libavutil/dict.h"
yading@10: #include "libavutil/xga_font_data.h"
yading@10: #include "libavutil/opt.h"
yading@10: #include "libavutil/timestamp.h"
yading@10: #include "audio.h"
yading@10: #include "avfilter.h"
yading@10: #include "formats.h"
yading@10: #include "internal.h"
yading@10: 
yading@10: #define MAX_CHANNELS 63
yading@10: 
yading@10: /* pre-filter coefficients */
yading@10: #define PRE_B0  1.53512485958697
yading@10: #define PRE_B1 -2.69169618940638
yading@10: #define PRE_B2  1.19839281085285
yading@10: #define PRE_A1 -1.69065929318241
yading@10: #define PRE_A2  0.73248077421585
yading@10: 
yading@10: /* RLB-filter coefficients */
yading@10: #define RLB_B0  1.0
yading@10: #define RLB_B1 -2.0
yading@10: #define RLB_B2  1.0
yading@10: #define RLB_A1 -1.99004745483398
yading@10: #define RLB_A2  0.99007225036621
yading@10: 
yading@10: #define ABS_THRES    -70            ///< silence gate: we discard anything below this absolute (LUFS) threshold
yading@10: #define ABS_UP_THRES  10            ///< upper loud limit to consider (ABS_THRES being the minimum)
yading@10: #define HIST_GRAIN   100            ///< defines histogram precision
yading@10: #define HIST_SIZE  ((ABS_UP_THRES - ABS_THRES) * HIST_GRAIN + 1)
yading@10: 
yading@10: /**
yading@10:  * An histogram is an array of HIST_SIZE hist_entry storing all the energies
yading@10:  * recorded (with an accuracy of 1/HIST_GRAIN) of the loudnesses from ABS_THRES
yading@10:  * (at 0) to ABS_UP_THRES (at HIST_SIZE-1).
yading@10:  * This fixed-size system avoids the need of a list of energies growing
yading@10:  * infinitely over the time and is thus more scalable.
yading@10:  */
yading@10: struct hist_entry {
yading@10:     int count;                      ///< how many times the corresponding value occurred
yading@10:     double energy;                  ///< E = 10^((L + 0.691) / 10)
yading@10:     double loudness;                ///< L = -0.691 + 10 * log10(E)
yading@10: };
yading@10: 
yading@10: struct integrator {
yading@10:     double *cache[MAX_CHANNELS];    ///< window of filtered samples (N ms)
yading@10:     int cache_pos;                  ///< focus on the last added bin in the cache array
yading@10:     double sum[MAX_CHANNELS];       ///< sum of the last N ms filtered samples (cache content)
yading@10:     int filled;                     ///< 1 if the cache is completely filled, 0 otherwise
yading@10:     double rel_threshold;           ///< relative threshold
yading@10:     double sum_kept_powers;         ///< sum of the powers (weighted sums) above absolute threshold
yading@10:     int nb_kept_powers;             ///< number of sum above absolute threshold
yading@10:     struct hist_entry *histogram;   ///< histogram of the powers, used to compute LRA and I
yading@10: };
yading@10: 
yading@10: struct rect { int x, y, w, h; };
yading@10: 
yading@10: typedef struct {
yading@10:     const AVClass *class;           ///< AVClass context for log and options purpose
yading@10: 
yading@10:     /* video  */
yading@10:     int do_video;                   ///< 1 if video output enabled, 0 otherwise
yading@10:     int w, h;                       ///< size of the video output
yading@10:     struct rect text;               ///< rectangle for the LU legend on the left
yading@10:     struct rect graph;              ///< rectangle for the main graph in the center
yading@10:     struct rect gauge;              ///< rectangle for the gauge on the right
yading@10:     AVFrame *outpicref;             ///< output picture reference, updated regularly
yading@10:     int meter;                      ///< select a EBU mode between +9 and +18
yading@10:     int scale_range;                ///< the range of LU values according to the meter
yading@10:     int y_zero_lu;                  ///< the y value (pixel position) for 0 LU
yading@10:     int *y_line_ref;                ///< y reference values for drawing the LU lines in the graph and the gauge
yading@10: 
yading@10:     /* audio */
yading@10:     int nb_channels;                ///< number of channels in the input
yading@10:     double *ch_weighting;           ///< channel weighting mapping
yading@10:     int sample_count;               ///< sample count used for refresh frequency, reset at refresh
yading@10: 
yading@10:     /* Filter caches.
yading@10:      * The mult by 3 in the following is for X[i], X[i-1] and X[i-2] */
yading@10:     double x[MAX_CHANNELS * 3];     ///< 3 input samples cache for each channel
yading@10:     double y[MAX_CHANNELS * 3];     ///< 3 pre-filter samples cache for each channel
yading@10:     double z[MAX_CHANNELS * 3];     ///< 3 RLB-filter samples cache for each channel
yading@10: 
yading@10: #define I400_BINS  (48000 * 4 / 10)
yading@10: #define I3000_BINS (48000 * 3)
yading@10:     struct integrator i400;         ///< 400ms integrator, used for Momentary loudness  (M), and Integrated loudness (I)
yading@10:     struct integrator i3000;        ///<    3s integrator, used for Short term loudness (S), and Loudness Range      (LRA)
yading@10: 
yading@10:     /* I and LRA specific */
yading@10:     double integrated_loudness;     ///< integrated loudness in LUFS (I)
yading@10:     double loudness_range;          ///< loudness range in LU (LRA)
yading@10:     double lra_low, lra_high;       ///< low and high LRA values
yading@10: 
yading@10:     /* misc */
yading@10:     int loglevel;                   ///< log level for frame logging
yading@10:     int metadata;                   ///< whether or not to inject loudness results in frames
yading@10: } EBUR128Context;
yading@10: 
yading@10: #define OFFSET(x) offsetof(EBUR128Context, x)
yading@10: #define A AV_OPT_FLAG_AUDIO_PARAM
yading@10: #define V AV_OPT_FLAG_VIDEO_PARAM
yading@10: #define F AV_OPT_FLAG_FILTERING_PARAM
yading@10: static const AVOption ebur128_options[] = {
yading@10:     { "video", "set video output", OFFSET(do_video), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, V|F },
yading@10:     { "size",  "set video size",   OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "640x480"}, 0, 0, V|F },
yading@10:     { "meter", "set scale meter (+9 to +18)",  OFFSET(meter), AV_OPT_TYPE_INT, {.i64 = 9}, 9, 18, V|F },
yading@10:     { "framelog", "force frame logging level", OFFSET(loglevel), AV_OPT_TYPE_INT, {.i64 = -1},   INT_MIN, INT_MAX, A|V|F, "level" },
yading@10:         { "info",    "information logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_INFO},    INT_MIN, INT_MAX, A|V|F, "level" },
yading@10:         { "verbose", "verbose logging level",     0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_VERBOSE}, INT_MIN, INT_MAX, A|V|F, "level" },
yading@10:     { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, A|V|F },
yading@10:     { NULL },
yading@10: };
yading@10: 
yading@10: AVFILTER_DEFINE_CLASS(ebur128);
yading@10: 
yading@10: static const uint8_t graph_colors[] = {
yading@10:     0xdd, 0x66, 0x66,   // value above 0LU non reached
yading@10:     0x66, 0x66, 0xdd,   // value below 0LU non reached
yading@10:     0x96, 0x33, 0x33,   // value above 0LU reached
yading@10:     0x33, 0x33, 0x96,   // value below 0LU reached
yading@10:     0xdd, 0x96, 0x96,   // value above 0LU line non reached
yading@10:     0x96, 0x96, 0xdd,   // value below 0LU line non reached
yading@10:     0xdd, 0x33, 0x33,   // value above 0LU line reached
yading@10:     0x33, 0x33, 0xdd,   // value below 0LU line reached
yading@10: };
yading@10: 
yading@10: static const uint8_t *get_graph_color(const EBUR128Context *ebur128, int v, int y)
yading@10: {
yading@10:     const int below0  = y > ebur128->y_zero_lu;
yading@10:     const int reached = y >= v;
yading@10:     const int line    = ebur128->y_line_ref[y] || y == ebur128->y_zero_lu;
yading@10:     const int colorid = 4*line + 2*reached + below0;
yading@10:     return graph_colors + 3*colorid;
yading@10: }
yading@10: 
yading@10: static inline int lu_to_y(const EBUR128Context *ebur128, double v)
yading@10: {
yading@10:     v += 2 * ebur128->meter;                            // make it in range [0;...]
yading@10:     v  = av_clipf(v, 0, ebur128->scale_range);          // make sure it's in the graph scale
yading@10:     v  = ebur128->scale_range - v;                      // invert value (y=0 is on top)
yading@10:     return v * ebur128->graph.h / ebur128->scale_range; // rescale from scale range to px height
yading@10: }
yading@10: 
yading@10: #define FONT8   0
yading@10: #define FONT16  1
yading@10: 
yading@10: static const uint8_t font_colors[] = {
yading@10:     0xdd, 0xdd, 0x00,
yading@10:     0x00, 0x96, 0x96,
yading@10: };
yading@10: 
yading@10: static void drawtext(AVFrame *pic, int x, int y, int ftid, const uint8_t *color, const char *fmt, ...)
yading@10: {
yading@10:     int i;
yading@10:     char buf[128] = {0};
yading@10:     const uint8_t *font;
yading@10:     int font_height;
yading@10:     va_list vl;
yading@10: 
yading@10:     if      (ftid == FONT16) font = avpriv_vga16_font, font_height = 16;
yading@10:     else if (ftid == FONT8)  font = avpriv_cga_font,   font_height =  8;
yading@10:     else return;
yading@10: 
yading@10:     va_start(vl, fmt);
yading@10:     vsnprintf(buf, sizeof(buf), fmt, vl);
yading@10:     va_end(vl);
yading@10: 
yading@10:     for (i = 0; buf[i]; i++) {
yading@10:         int char_y, mask;
yading@10:         uint8_t *p = pic->data[0] + y*pic->linesize[0] + (x + i*8)*3;
yading@10: 
yading@10:         for (char_y = 0; char_y < font_height; char_y++) {
yading@10:             for (mask = 0x80; mask; mask >>= 1) {
yading@10:                 if (font[buf[i] * font_height + char_y] & mask)
yading@10:                     memcpy(p, color, 3);
yading@10:                 else
yading@10:                     memcpy(p, "\x00\x00\x00", 3);
yading@10:                 p += 3;
yading@10:             }
yading@10:             p += pic->linesize[0] - 8*3;
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void drawline(AVFrame *pic, int x, int y, int len, int step)
yading@10: {
yading@10:     int i;
yading@10:     uint8_t *p = pic->data[0] + y*pic->linesize[0] + x*3;
yading@10: 
yading@10:     for (i = 0; i < len; i++) {
yading@10:         memcpy(p, "\x00\xff\x00", 3);
yading@10:         p += step;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static int config_video_output(AVFilterLink *outlink)
yading@10: {
yading@10:     int i, x, y;
yading@10:     uint8_t *p;
yading@10:     AVFilterContext *ctx = outlink->src;
yading@10:     EBUR128Context *ebur128 = ctx->priv;
yading@10:     AVFrame *outpicref;
yading@10: 
yading@10:     /* check if there is enough space to represent everything decently */
yading@10:     if (ebur128->w < 640 || ebur128->h < 480) {
yading@10:         av_log(ctx, AV_LOG_ERROR, "Video size %dx%d is too small, "
yading@10:                "minimum size is 640x480\n", ebur128->w, ebur128->h);
yading@10:         return AVERROR(EINVAL);
yading@10:     }
yading@10:     outlink->w = ebur128->w;
yading@10:     outlink->h = ebur128->h;
yading@10: 
yading@10: #define PAD 8
yading@10: 
yading@10:     /* configure text area position and size */
yading@10:     ebur128->text.x  = PAD;
yading@10:     ebur128->text.y  = 40;
yading@10:     ebur128->text.w  = 3 * 8;   // 3 characters
yading@10:     ebur128->text.h  = ebur128->h - PAD - ebur128->text.y;
yading@10: 
yading@10:     /* configure gauge position and size */
yading@10:     ebur128->gauge.w = 20;
yading@10:     ebur128->gauge.h = ebur128->text.h;
yading@10:     ebur128->gauge.x = ebur128->w - PAD - ebur128->gauge.w;
yading@10:     ebur128->gauge.y = ebur128->text.y;
yading@10: 
yading@10:     /* configure graph position and size */
yading@10:     ebur128->graph.x = ebur128->text.x + ebur128->text.w + PAD;
yading@10:     ebur128->graph.y = ebur128->gauge.y;
yading@10:     ebur128->graph.w = ebur128->gauge.x - ebur128->graph.x - PAD;
yading@10:     ebur128->graph.h = ebur128->gauge.h;
yading@10: 
yading@10:     /* graph and gauge share the LU-to-pixel code */
yading@10:     av_assert0(ebur128->graph.h == ebur128->gauge.h);
yading@10: 
yading@10:     /* prepare the initial picref buffer */
yading@10:     av_frame_free(&ebur128->outpicref);
yading@10:     ebur128->outpicref = outpicref =
yading@10:         ff_get_video_buffer(outlink, outlink->w, outlink->h);
yading@10:     if (!outpicref)
yading@10:         return AVERROR(ENOMEM);
yading@10:     outlink->sample_aspect_ratio = (AVRational){1,1};
yading@10: 
yading@10:     /* init y references values (to draw LU lines) */
yading@10:     ebur128->y_line_ref = av_calloc(ebur128->graph.h + 1, sizeof(*ebur128->y_line_ref));
yading@10:     if (!ebur128->y_line_ref)
yading@10:         return AVERROR(ENOMEM);
yading@10: 
yading@10:     /* black background */
yading@10:     memset(outpicref->data[0], 0, ebur128->h * outpicref->linesize[0]);
yading@10: 
yading@10:     /* draw LU legends */
yading@10:     drawtext(outpicref, PAD, PAD+16, FONT8, font_colors+3, " LU");
yading@10:     for (i = ebur128->meter; i >= -ebur128->meter * 2; i--) {
yading@10:         y = lu_to_y(ebur128, i);
yading@10:         x = PAD + (i < 10 && i > -10) * 8;
yading@10:         ebur128->y_line_ref[y] = i;
yading@10:         y -= 4; // -4 to center vertically
yading@10:         drawtext(outpicref, x, y + ebur128->graph.y, FONT8, font_colors+3,
yading@10:                  "%c%d", i < 0 ? '-' : i > 0 ? '+' : ' ', FFABS(i));
yading@10:     }
yading@10: 
yading@10:     /* draw graph */
yading@10:     ebur128->y_zero_lu = lu_to_y(ebur128, 0);
yading@10:     p = outpicref->data[0] + ebur128->graph.y * outpicref->linesize[0]
yading@10:                            + ebur128->graph.x * 3;
yading@10:     for (y = 0; y < ebur128->graph.h; y++) {
yading@10:         const uint8_t *c = get_graph_color(ebur128, INT_MAX, y);
yading@10: 
yading@10:         for (x = 0; x < ebur128->graph.w; x++)
yading@10:             memcpy(p + x*3, c, 3);
yading@10:         p += outpicref->linesize[0];
yading@10:     }
yading@10: 
yading@10:     /* draw fancy rectangles around the graph and the gauge */
yading@10: #define DRAW_RECT(r) do { \
yading@10:     drawline(outpicref, r.x,       r.y - 1,   r.w, 3); \
yading@10:     drawline(outpicref, r.x,       r.y + r.h, r.w, 3); \
yading@10:     drawline(outpicref, r.x - 1,   r.y,       r.h, outpicref->linesize[0]); \
yading@10:     drawline(outpicref, r.x + r.w, r.y,       r.h, outpicref->linesize[0]); \
yading@10: } while (0)
yading@10:     DRAW_RECT(ebur128->graph);
yading@10:     DRAW_RECT(ebur128->gauge);
yading@10: 
yading@10:     outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static int config_audio_input(AVFilterLink *inlink)
yading@10: {
yading@10:     AVFilterContext *ctx = inlink->dst;
yading@10:     EBUR128Context *ebur128 = ctx->priv;
yading@10: 
yading@10:     /* force 100ms framing in case of metadata injection: the frames must have
yading@10:      * a granularity of the window overlap to be accurately exploited */
yading@10:     if (ebur128->metadata)
yading@10:         inlink->min_samples =
yading@10:         inlink->max_samples =
yading@10:         inlink->partial_buf_size = inlink->sample_rate / 10;
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static int config_audio_output(AVFilterLink *outlink)
yading@10: {
yading@10:     int i;
yading@10:     int idx_bitposn = 0;
yading@10:     AVFilterContext *ctx = outlink->src;
yading@10:     EBUR128Context *ebur128 = ctx->priv;
yading@10:     const int nb_channels = av_get_channel_layout_nb_channels(outlink->channel_layout);
yading@10: 
yading@10: #define BACK_MASK (AV_CH_BACK_LEFT    |AV_CH_BACK_CENTER    |AV_CH_BACK_RIGHT| \
yading@10:                    AV_CH_TOP_BACK_LEFT|AV_CH_TOP_BACK_CENTER|AV_CH_TOP_BACK_RIGHT| \
yading@10:                    AV_CH_SIDE_LEFT                          |AV_CH_SIDE_RIGHT| \
yading@10:                    AV_CH_SURROUND_DIRECT_LEFT               |AV_CH_SURROUND_DIRECT_RIGHT)
yading@10: 
yading@10:     ebur128->nb_channels  = nb_channels;
yading@10:     ebur128->ch_weighting = av_calloc(nb_channels, sizeof(*ebur128->ch_weighting));
yading@10:     if (!ebur128->ch_weighting)
yading@10:         return AVERROR(ENOMEM);
yading@10: 
yading@10:     for (i = 0; i < nb_channels; i++) {
yading@10: 
yading@10:         /* find the next bit that is set starting from the right */
yading@10:         while ((outlink->channel_layout & 1ULL<<idx_bitposn) == 0 && idx_bitposn < 63)
yading@10:             idx_bitposn++;
yading@10: 
yading@10:         /* channel weighting */
yading@10:         if ((1ULL<<idx_bitposn & AV_CH_LOW_FREQUENCY) ||
yading@10:             (1ULL<<idx_bitposn & AV_CH_LOW_FREQUENCY_2)) {
yading@10:             ebur128->ch_weighting[i] = 0;
yading@10:         } else if (1ULL<<idx_bitposn & BACK_MASK) {
yading@10:             ebur128->ch_weighting[i] = 1.41;
yading@10:         } else {
yading@10:             ebur128->ch_weighting[i] = 1.0;
yading@10:         }
yading@10: 
yading@10:         idx_bitposn++;
yading@10: 
yading@10:         if (!ebur128->ch_weighting[i])
yading@10:             continue;
yading@10: 
yading@10:         /* bins buffer for the two integration window (400ms and 3s) */
yading@10:         ebur128->i400.cache[i]  = av_calloc(I400_BINS,  sizeof(*ebur128->i400.cache[0]));
yading@10:         ebur128->i3000.cache[i] = av_calloc(I3000_BINS, sizeof(*ebur128->i3000.cache[0]));
yading@10:         if (!ebur128->i400.cache[i] || !ebur128->i3000.cache[i])
yading@10:             return AVERROR(ENOMEM);
yading@10:     }
yading@10: 
yading@10:     outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: #define ENERGY(loudness) (pow(10, ((loudness) + 0.691) / 10.))
yading@10: #define LOUDNESS(energy) (-0.691 + 10 * log10(energy))
yading@10: 
yading@10: static struct hist_entry *get_histogram(void)
yading@10: {
yading@10:     int i;
yading@10:     struct hist_entry *h = av_calloc(HIST_SIZE, sizeof(*h));
yading@10: 
yading@10:     if (!h)
yading@10:         return NULL;
yading@10:     for (i = 0; i < HIST_SIZE; i++) {
yading@10:         h[i].loudness = i / (double)HIST_GRAIN + ABS_THRES;
yading@10:         h[i].energy   = ENERGY(h[i].loudness);
yading@10:     }
yading@10:     return h;
yading@10: }
yading@10: 
yading@10: static av_cold int init(AVFilterContext *ctx)
yading@10: {
yading@10:     EBUR128Context *ebur128 = ctx->priv;
yading@10:     AVFilterPad pad;
yading@10: 
yading@10:     if (ebur128->loglevel != AV_LOG_INFO &&
yading@10:         ebur128->loglevel != AV_LOG_VERBOSE) {
yading@10:         if (ebur128->do_video || ebur128->metadata)
yading@10:             ebur128->loglevel = AV_LOG_VERBOSE;
yading@10:         else
yading@10:             ebur128->loglevel = AV_LOG_INFO;
yading@10:     }
yading@10: 
yading@10:     // if meter is  +9 scale, scale range is from -18 LU to  +9 LU (or 3*9)
yading@10:     // if meter is +18 scale, scale range is from -36 LU to +18 LU (or 3*18)
yading@10:     ebur128->scale_range = 3 * ebur128->meter;
yading@10: 
yading@10:     ebur128->i400.histogram  = get_histogram();
yading@10:     ebur128->i3000.histogram = get_histogram();
yading@10:     if (!ebur128->i400.histogram || !ebur128->i3000.histogram)
yading@10:         return AVERROR(ENOMEM);
yading@10: 
yading@10:     ebur128->integrated_loudness = ABS_THRES;
yading@10:     ebur128->loudness_range = 0;
yading@10: 
yading@10:     /* insert output pads */
yading@10:     if (ebur128->do_video) {
yading@10:         pad = (AVFilterPad){
yading@10:             .name         = av_strdup("out0"),
yading@10:             .type         = AVMEDIA_TYPE_VIDEO,
yading@10:             .config_props = config_video_output,
yading@10:         };
yading@10:         if (!pad.name)
yading@10:             return AVERROR(ENOMEM);
yading@10:         ff_insert_outpad(ctx, 0, &pad);
yading@10:     }
yading@10:     pad = (AVFilterPad){
yading@10:         .name         = av_asprintf("out%d", ebur128->do_video),
yading@10:         .type         = AVMEDIA_TYPE_AUDIO,
yading@10:         .config_props = config_audio_output,
yading@10:     };
yading@10:     if (!pad.name)
yading@10:         return AVERROR(ENOMEM);
yading@10:     ff_insert_outpad(ctx, ebur128->do_video, &pad);
yading@10: 
yading@10:     /* summary */
yading@10:     av_log(ctx, AV_LOG_VERBOSE, "EBU +%d scale\n", ebur128->meter);
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: #define HIST_POS(power) (int)(((power) - ABS_THRES) * HIST_GRAIN)
yading@10: 
yading@10: /* loudness and power should be set such as loudness = -0.691 +
yading@10:  * 10*log10(power), we just avoid doing that calculus two times */
yading@10: static int gate_update(struct integrator *integ, double power,
yading@10:                        double loudness, int gate_thres)
yading@10: {
yading@10:     int ipower;
yading@10:     double relative_threshold;
yading@10:     int gate_hist_pos;
yading@10: 
yading@10:     /* update powers histograms by incrementing current power count */
yading@10:     ipower = av_clip(HIST_POS(loudness), 0, HIST_SIZE - 1);
yading@10:     integ->histogram[ipower].count++;
yading@10: 
yading@10:     /* compute relative threshold and get its position in the histogram */
yading@10:     integ->sum_kept_powers += power;
yading@10:     integ->nb_kept_powers++;
yading@10:     relative_threshold = integ->sum_kept_powers / integ->nb_kept_powers;
yading@10:     if (!relative_threshold)
yading@10:         relative_threshold = 1e-12;
yading@10:     integ->rel_threshold = LOUDNESS(relative_threshold) + gate_thres;
yading@10:     gate_hist_pos = av_clip(HIST_POS(integ->rel_threshold), 0, HIST_SIZE - 1);
yading@10: 
yading@10:     return gate_hist_pos;
yading@10: }
yading@10: 
yading@10: static int filter_frame(AVFilterLink *inlink, AVFrame *insamples)
yading@10: {
yading@10:     int i, ch, idx_insample;
yading@10:     AVFilterContext *ctx = inlink->dst;
yading@10:     EBUR128Context *ebur128 = ctx->priv;
yading@10:     const int nb_channels = ebur128->nb_channels;
yading@10:     const int nb_samples  = insamples->nb_samples;
yading@10:     const double *samples = (double *)insamples->data[0];
yading@10:     AVFrame *pic = ebur128->outpicref;
yading@10: 
yading@10:     for (idx_insample = 0; idx_insample < nb_samples; idx_insample++) {
yading@10:         const int bin_id_400  = ebur128->i400.cache_pos;
yading@10:         const int bin_id_3000 = ebur128->i3000.cache_pos;
yading@10: 
yading@10: #define MOVE_TO_NEXT_CACHED_ENTRY(time) do {                \
yading@10:     ebur128->i##time.cache_pos++;                           \
yading@10:     if (ebur128->i##time.cache_pos == I##time##_BINS) {     \
yading@10:         ebur128->i##time.filled    = 1;                     \
yading@10:         ebur128->i##time.cache_pos = 0;                     \
yading@10:     }                                                       \
yading@10: } while (0)
yading@10: 
yading@10:         MOVE_TO_NEXT_CACHED_ENTRY(400);
yading@10:         MOVE_TO_NEXT_CACHED_ENTRY(3000);
yading@10: 
yading@10:         for (ch = 0; ch < nb_channels; ch++) {
yading@10:             double bin;
yading@10: 
yading@10:             ebur128->x[ch * 3] = *samples++; // set X[i]
yading@10: 
yading@10:             if (!ebur128->ch_weighting[ch])
yading@10:                 continue;
yading@10: 
yading@10:             /* Y[i] = X[i]*b0 + X[i-1]*b1 + X[i-2]*b2 - Y[i-1]*a1 - Y[i-2]*a2 */
yading@10: #define FILTER(Y, X, name) do {                                                 \
yading@10:             double *dst = ebur128->Y + ch*3;                                    \
yading@10:             double *src = ebur128->X + ch*3;                                    \
yading@10:             dst[2] = dst[1];                                                    \
yading@10:             dst[1] = dst[0];                                                    \
yading@10:             dst[0] = src[0]*name##_B0 + src[1]*name##_B1 + src[2]*name##_B2     \
yading@10:                                       - dst[1]*name##_A1 - dst[2]*name##_A2;    \
yading@10: } while (0)
yading@10: 
yading@10:             // TODO: merge both filters in one?
yading@10:             FILTER(y, x, PRE);  // apply pre-filter
yading@10:             ebur128->x[ch * 3 + 2] = ebur128->x[ch * 3 + 1];
yading@10:             ebur128->x[ch * 3 + 1] = ebur128->x[ch * 3    ];
yading@10:             FILTER(z, y, RLB);  // apply RLB-filter
yading@10: 
yading@10:             bin = ebur128->z[ch * 3] * ebur128->z[ch * 3];
yading@10: 
yading@10:             /* add the new value, and limit the sum to the cache size (400ms or 3s)
yading@10:              * by removing the oldest one */
yading@10:             ebur128->i400.sum [ch] = ebur128->i400.sum [ch] + bin - ebur128->i400.cache [ch][bin_id_400];
yading@10:             ebur128->i3000.sum[ch] = ebur128->i3000.sum[ch] + bin - ebur128->i3000.cache[ch][bin_id_3000];
yading@10: 
yading@10:             /* override old cache entry with the new value */
yading@10:             ebur128->i400.cache [ch][bin_id_400 ] = bin;
yading@10:             ebur128->i3000.cache[ch][bin_id_3000] = bin;
yading@10:         }
yading@10: 
yading@10:         /* For integrated loudness, gating blocks are 400ms long with 75%
yading@10:          * overlap (see BS.1770-2 p5), so a re-computation is needed each 100ms
yading@10:          * (4800 samples at 48kHz). */
yading@10:         if (++ebur128->sample_count == 4800) {
yading@10:             double loudness_400, loudness_3000;
yading@10:             double power_400 = 1e-12, power_3000 = 1e-12;
yading@10:             AVFilterLink *outlink = ctx->outputs[0];
yading@10:             const int64_t pts = insamples->pts +
yading@10:                 av_rescale_q(idx_insample, (AVRational){ 1, inlink->sample_rate },
yading@10:                              outlink->time_base);
yading@10: 
yading@10:             ebur128->sample_count = 0;
yading@10: 
yading@10: #define COMPUTE_LOUDNESS(m, time) do {                                              \
yading@10:     if (ebur128->i##time.filled) {                                                  \
yading@10:         /* weighting sum of the last <time> ms */                                   \
yading@10:         for (ch = 0; ch < nb_channels; ch++)                                        \
yading@10:             power_##time += ebur128->ch_weighting[ch] * ebur128->i##time.sum[ch];   \
yading@10:         power_##time /= I##time##_BINS;                                             \
yading@10:     }                                                                               \
yading@10:     loudness_##time = LOUDNESS(power_##time);                                       \
yading@10: } while (0)
yading@10: 
yading@10:             COMPUTE_LOUDNESS(M,  400);
yading@10:             COMPUTE_LOUDNESS(S, 3000);
yading@10: 
yading@10:             /* Integrated loudness */
yading@10: #define I_GATE_THRES -10  // initially defined to -8 LU in the first EBU standard
yading@10: 
yading@10:             if (loudness_400 >= ABS_THRES) {
yading@10:                 double integrated_sum = 0;
yading@10:                 int nb_integrated = 0;
yading@10:                 int gate_hist_pos = gate_update(&ebur128->i400, power_400,
yading@10:                                                 loudness_400, I_GATE_THRES);
yading@10: 
yading@10:                 /* compute integrated loudness by summing the histogram values
yading@10:                  * above the relative threshold */
yading@10:                 for (i = gate_hist_pos; i < HIST_SIZE; i++) {
yading@10:                     const int nb_v = ebur128->i400.histogram[i].count;
yading@10:                     nb_integrated  += nb_v;
yading@10:                     integrated_sum += nb_v * ebur128->i400.histogram[i].energy;
yading@10:                 }
yading@10:                 if (nb_integrated)
yading@10:                     ebur128->integrated_loudness = LOUDNESS(integrated_sum / nb_integrated);
yading@10:             }
yading@10: 
yading@10:             /* LRA */
yading@10: #define LRA_GATE_THRES -20
yading@10: #define LRA_LOWER_PRC   10
yading@10: #define LRA_HIGHER_PRC  95
yading@10: 
yading@10:             /* XXX: example code in EBU 3342 is ">=" but formula in BS.1770
yading@10:              * specs is ">" */
yading@10:             if (loudness_3000 >= ABS_THRES) {
yading@10:                 int nb_powers = 0;
yading@10:                 int gate_hist_pos = gate_update(&ebur128->i3000, power_3000,
yading@10:                                                 loudness_3000, LRA_GATE_THRES);
yading@10: 
yading@10:                 for (i = gate_hist_pos; i < HIST_SIZE; i++)
yading@10:                     nb_powers += ebur128->i3000.histogram[i].count;
yading@10:                 if (nb_powers) {
yading@10:                     int n, nb_pow;
yading@10: 
yading@10:                     /* get lower loudness to consider */
yading@10:                     n = 0;
yading@10:                     nb_pow = LRA_LOWER_PRC  * nb_powers / 100. + 0.5;
yading@10:                     for (i = gate_hist_pos; i < HIST_SIZE; i++) {
yading@10:                         n += ebur128->i3000.histogram[i].count;
yading@10:                         if (n >= nb_pow) {
yading@10:                             ebur128->lra_low = ebur128->i3000.histogram[i].loudness;
yading@10:                             break;
yading@10:                         }
yading@10:                     }
yading@10: 
yading@10:                     /* get higher loudness to consider */
yading@10:                     n = nb_powers;
yading@10:                     nb_pow = LRA_HIGHER_PRC * nb_powers / 100. + 0.5;
yading@10:                     for (i = HIST_SIZE - 1; i >= 0; i--) {
yading@10:                         n -= ebur128->i3000.histogram[i].count;
yading@10:                         if (n < nb_pow) {
yading@10:                             ebur128->lra_high = ebur128->i3000.histogram[i].loudness;
yading@10:                             break;
yading@10:                         }
yading@10:                     }
yading@10: 
yading@10:                     // XXX: show low & high on the graph?
yading@10:                     ebur128->loudness_range = ebur128->lra_high - ebur128->lra_low;
yading@10:                 }
yading@10:             }
yading@10: 
yading@10: #define LOG_FMT "M:%6.1f S:%6.1f     I:%6.1f LUFS     LRA:%6.1f LU"
yading@10: 
yading@10:             /* push one video frame */
yading@10:             if (ebur128->do_video) {
yading@10:                 int x, y, ret;
yading@10:                 uint8_t *p;
yading@10: 
yading@10:                 const int y_loudness_lu_graph = lu_to_y(ebur128, loudness_3000 + 23);
yading@10:                 const int y_loudness_lu_gauge = lu_to_y(ebur128, loudness_400  + 23);
yading@10: 
yading@10:                 /* draw the graph using the short-term loudness */
yading@10:                 p = pic->data[0] + ebur128->graph.y*pic->linesize[0] + ebur128->graph.x*3;
yading@10:                 for (y = 0; y < ebur128->graph.h; y++) {
yading@10:                     const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_graph, y);
yading@10: 
yading@10:                     memmove(p, p + 3, (ebur128->graph.w - 1) * 3);
yading@10:                     memcpy(p + (ebur128->graph.w - 1) * 3, c, 3);
yading@10:                     p += pic->linesize[0];
yading@10:                 }
yading@10: 
yading@10:                 /* draw the gauge using the momentary loudness */
yading@10:                 p = pic->data[0] + ebur128->gauge.y*pic->linesize[0] + ebur128->gauge.x*3;
yading@10:                 for (y = 0; y < ebur128->gauge.h; y++) {
yading@10:                     const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_gauge, y);
yading@10: 
yading@10:                     for (x = 0; x < ebur128->gauge.w; x++)
yading@10:                         memcpy(p + x*3, c, 3);
yading@10:                     p += pic->linesize[0];
yading@10:                 }
yading@10: 
yading@10:                 /* draw textual info */
yading@10:                 drawtext(pic, PAD, PAD - PAD/2, FONT16, font_colors,
yading@10:                          LOG_FMT "     ", // padding to erase trailing characters
yading@10:                          loudness_400, loudness_3000,
yading@10:                          ebur128->integrated_loudness, ebur128->loudness_range);
yading@10: 
yading@10:                 /* set pts and push frame */
yading@10:                 pic->pts = pts;
yading@10:                 ret = ff_filter_frame(outlink, av_frame_clone(pic));
yading@10:                 if (ret < 0)
yading@10:                     return ret;
yading@10:             }
yading@10: 
yading@10:             if (ebur128->metadata) { /* happens only once per filter_frame call */
yading@10:                 char metabuf[128];
yading@10: #define SET_META(name, var) do {                                            \
yading@10:     snprintf(metabuf, sizeof(metabuf), "%.3f", var);                        \
yading@10:     av_dict_set(&insamples->metadata, "lavfi.r128." name, metabuf, 0);      \
yading@10: } while (0)
yading@10:                 SET_META("M",        loudness_400);
yading@10:                 SET_META("S",        loudness_3000);
yading@10:                 SET_META("I",        ebur128->integrated_loudness);
yading@10:                 SET_META("LRA",      ebur128->loudness_range);
yading@10:                 SET_META("LRA.low",  ebur128->lra_low);
yading@10:                 SET_META("LRA.high", ebur128->lra_high);
yading@10:             }
yading@10: 
yading@10:             av_log(ctx, ebur128->loglevel, "t: %-10s " LOG_FMT "\n",
yading@10:                    av_ts2timestr(pts, &outlink->time_base),
yading@10:                    loudness_400, loudness_3000,
yading@10:                    ebur128->integrated_loudness, ebur128->loudness_range);
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     return ff_filter_frame(ctx->outputs[ebur128->do_video], insamples);
yading@10: }
yading@10: 
yading@10: static int query_formats(AVFilterContext *ctx)
yading@10: {
yading@10:     EBUR128Context *ebur128 = ctx->priv;
yading@10:     AVFilterFormats *formats;
yading@10:     AVFilterChannelLayouts *layouts;
yading@10:     AVFilterLink *inlink = ctx->inputs[0];
yading@10:     AVFilterLink *outlink = ctx->outputs[0];
yading@10: 
yading@10:     static const enum AVSampleFormat sample_fmts[] = { AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_NONE };
yading@10:     static const int input_srate[] = {48000, -1}; // ITU-R BS.1770 provides coeff only for 48kHz
yading@10:     static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_RGB24, AV_PIX_FMT_NONE };
yading@10: 
yading@10:     /* set optional output video format */
yading@10:     if (ebur128->do_video) {
yading@10:         formats = ff_make_format_list(pix_fmts);
yading@10:         if (!formats)
yading@10:             return AVERROR(ENOMEM);
yading@10:         ff_formats_ref(formats, &outlink->in_formats);
yading@10:         outlink = ctx->outputs[1];
yading@10:     }
yading@10: 
yading@10:     /* set input and output audio formats
yading@10:      * Note: ff_set_common_* functions are not used because they affect all the
yading@10:      * links, and thus break the video format negociation */
yading@10:     formats = ff_make_format_list(sample_fmts);
yading@10:     if (!formats)
yading@10:         return AVERROR(ENOMEM);
yading@10:     ff_formats_ref(formats, &inlink->out_formats);
yading@10:     ff_formats_ref(formats, &outlink->in_formats);
yading@10: 
yading@10:     layouts = ff_all_channel_layouts();
yading@10:     if (!layouts)
yading@10:         return AVERROR(ENOMEM);
yading@10:     ff_channel_layouts_ref(layouts, &inlink->out_channel_layouts);
yading@10:     ff_channel_layouts_ref(layouts, &outlink->in_channel_layouts);
yading@10: 
yading@10:     formats = ff_make_format_list(input_srate);
yading@10:     if (!formats)
yading@10:         return AVERROR(ENOMEM);
yading@10:     ff_formats_ref(formats, &inlink->out_samplerates);
yading@10:     ff_formats_ref(formats, &outlink->in_samplerates);
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static av_cold void uninit(AVFilterContext *ctx)
yading@10: {
yading@10:     int i;
yading@10:     EBUR128Context *ebur128 = ctx->priv;
yading@10: 
yading@10:     av_log(ctx, AV_LOG_INFO, "Summary:\n\n"
yading@10:            "  Integrated loudness:\n"
yading@10:            "    I:         %5.1f LUFS\n"
yading@10:            "    Threshold: %5.1f LUFS\n\n"
yading@10:            "  Loudness range:\n"
yading@10:            "    LRA:       %5.1f LU\n"
yading@10:            "    Threshold: %5.1f LUFS\n"
yading@10:            "    LRA low:   %5.1f LUFS\n"
yading@10:            "    LRA high:  %5.1f LUFS\n",
yading@10:            ebur128->integrated_loudness, ebur128->i400.rel_threshold,
yading@10:            ebur128->loudness_range,      ebur128->i3000.rel_threshold,
yading@10:            ebur128->lra_low, ebur128->lra_high);
yading@10: 
yading@10:     av_freep(&ebur128->y_line_ref);
yading@10:     av_freep(&ebur128->ch_weighting);
yading@10:     av_freep(&ebur128->i400.histogram);
yading@10:     av_freep(&ebur128->i3000.histogram);
yading@10:     for (i = 0; i < ebur128->nb_channels; i++) {
yading@10:         av_freep(&ebur128->i400.cache[i]);
yading@10:         av_freep(&ebur128->i3000.cache[i]);
yading@10:     }
yading@10:     for (i = 0; i < ctx->nb_outputs; i++)
yading@10:         av_freep(&ctx->output_pads[i].name);
yading@10:     av_frame_free(&ebur128->outpicref);
yading@10: }
yading@10: 
yading@10: static const AVFilterPad ebur128_inputs[] = {
yading@10:     {
yading@10:         .name             = "default",
yading@10:         .type             = AVMEDIA_TYPE_AUDIO,
yading@10:         .get_audio_buffer = ff_null_get_audio_buffer,
yading@10:         .filter_frame     = filter_frame,
yading@10:         .config_props     = config_audio_input,
yading@10:     },
yading@10:     { NULL }
yading@10: };
yading@10: 
yading@10: AVFilter avfilter_af_ebur128 = {
yading@10:     .name          = "ebur128",
yading@10:     .description   = NULL_IF_CONFIG_SMALL("EBU R128 scanner."),
yading@10:     .priv_size     = sizeof(EBUR128Context),
yading@10:     .init          = init,
yading@10:     .uninit        = uninit,
yading@10:     .query_formats = query_formats,
yading@10:     .inputs        = ebur128_inputs,
yading@10:     .outputs       = NULL,
yading@10:     .priv_class    = &ebur128_class,
yading@10:     .flags         = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
yading@10: };