src/lib/audio_analyser.cc

   1 /*
   2     Copyright (C) 2021 Carl Hetherington <cth@carlh.net>
   3
   4     This file is part of DCP-o-matic.
   5
   6     DCP-o-matic is free software; you can redistribute it and/or modify
   7     it under the terms of the GNU General Public License as published by
   8     the Free Software Foundation; either version 2 of the License, or
   9     (at your option) any later version.
  10
  11     DCP-o-matic is distributed in the hope that it will be useful,
  12     but WITHOUT ANY WARRANTY; without even the implied warranty of
  13     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14     GNU General Public License for more details.
  15
  16     You should have received a copy of the GNU General Public License
  17     along with DCP-o-matic.  If not, see <http://www.gnu.org/licenses/>.
  18
  19 */
  20
  21
  22 #include "audio_analyser.h"
  23 #include "audio_analysis.h"
  24 #include "audio_buffers.h"
  25 #include "audio_content.h"
  26 #include "audio_filter_graph.h"
  27 #include "audio_point.h"
  28 #include "config.h"
  29 #include "dcpomatic_log.h"
  30 #include "film.h"
  31 #include "filter.h"
  32 #include "playlist.h"
  33 #include "types.h"
  34 #include "warnings.h"
  35 extern "C" {
  36 #include <leqm_nrt.h>
  37 DCPOMATIC_DISABLE_WARNINGS
  38 #include <libavutil/channel_layout.h>
  39 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  40 #include <libavfilter/f_ebur128.h>
  41 #endif
  42 DCPOMATIC_ENABLE_WARNINGS
  43 }
  44
  45
  46 using std::make_shared;
  47 using std::max;
  48 using std::shared_ptr;
  49 using std::vector;
  50 using namespace dcpomatic;
  51
  52
  53 static auto constexpr num_points = 1024;
  54
  55
  56 AudioAnalyser::AudioAnalyser (shared_ptr<const Film> film, shared_ptr<const Playlist> playlist, bool from_zero, std::function<void (float)> set_progress)
  57         : _film (film)
  58         , _playlist (playlist)
  59         , _set_progress (set_progress)
  60 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  61         , _ebur128 (new AudioFilterGraph(film->audio_frame_rate(), film->audio_channels()))
  62 #endif
  63         , _sample_peak (new float[film->audio_channels()])
  64         , _sample_peak_frame (new Frame[film->audio_channels()])
  65         , _analysis (film->audio_channels())
  66 {
  67
  68 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  69         _filters.push_back (new Filter("ebur128", "ebur128", "audio", "ebur128=peak=true"));
  70         _ebur128->setup (_filters);
  71 #endif
  72
  73         _current = new AudioPoint[_film->audio_channels()];
  74
  75         if (!from_zero) {
  76                 _start = _playlist->start().get_value_or(DCPTime());
  77         }
  78
  79         for (int i = 0; i < film->audio_channels(); ++i) {
  80                 _sample_peak[i] = 0;
  81                 _sample_peak_frame[i] = 0;
  82         }
  83
  84         auto add_if_required = [](vector<double>& v, size_t i, double db) {
  85                 if (v.size() > i) {
  86                         v[i] = pow(10, db / 20);
  87                 }
  88         };
  89
  90         /* XXX: is this right?  Especially for more than 5.1? */
  91         vector<double> channel_corrections(film->audio_channels(), 1);
  92         add_if_required (channel_corrections,  4,   -3); // Ls
  93         add_if_required (channel_corrections,  5,   -3); // Rs
  94         add_if_required (channel_corrections,  6, -144); // HI
  95         add_if_required (channel_corrections,  7, -144); // VI
  96         add_if_required (channel_corrections,  8,   -3); // Lc
  97         add_if_required (channel_corrections,  9,   -3); // Rc
  98         add_if_required (channel_corrections, 10,   -3); // Lc
  99         add_if_required (channel_corrections, 11,   -3); // Rc
 100         add_if_required (channel_corrections, 12, -144); // DBox
 101         add_if_required (channel_corrections, 13, -144); // Sync
 102         add_if_required (channel_corrections, 14, -144); // Sign Language
 103         add_if_required (channel_corrections, 15, -144); // Unused
 104
 105         _leqm.reset(new leqm_nrt::Calculator(
 106                 film->audio_channels(),
 107                 film->audio_frame_rate(),
 108                 24,
 109                 channel_corrections,
 110                 850, // suggested by leqm_nrt CLI source
 111                 64,  // suggested by leqm_nrt CLI source
 112                 boost::thread::hardware_concurrency()
 113                 ));
 114
 115         DCPTime const length = _playlist->length (_film);
 116
 117         Frame const len = DCPTime (length - _start).frames_round (film->audio_frame_rate());
 118         _samples_per_point = max (int64_t (1), len / num_points);
 119 }
 120
 121
 122 AudioAnalyser::~AudioAnalyser ()
 123 {
 124         delete[] _current;
 125         for (auto i: _filters) {
 126                 delete const_cast<Filter*> (i);
 127         }
 128         delete[] _sample_peak;
 129         delete[] _sample_peak_frame;
 130 }
 131
 132
 133 void
 134 AudioAnalyser::analyse (shared_ptr<const AudioBuffers> b, DCPTime time)
 135 {
 136         LOG_DEBUG_AUDIO_ANALYSIS("Received %1 frames at %2", b->frames(), to_string(time));
 137         DCPOMATIC_ASSERT (time >= _start);
 138
 139 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
 140         if (Config::instance()->analyse_ebur128 ()) {
 141                 _ebur128->process (b);
 142         }
 143 #endif
 144
 145         int const frames = b->frames ();
 146         int const channels = b->channels ();
 147         vector<double> interleaved(frames * channels);
 148
 149         for (int j = 0; j < channels; ++j) {
 150                 float* data = b->data(j);
 151                 for (int i = 0; i < frames; ++i) {
 152                         float s = data[i];
 153
 154                         interleaved[i * channels + j] = s;
 155
 156                         float as = fabsf (s);
 157                         if (as < 10e-7) {
 158                                 /* We may struggle to serialise and recover inf or -inf, so prevent such
 159                                    values by replacing with this (140dB down) */
 160                                 s = as = 10e-7;
 161                         }
 162                         _current[j][AudioPoint::RMS] += pow (s, 2);
 163                         _current[j][AudioPoint::PEAK] = max (_current[j][AudioPoint::PEAK], as);
 164
 165                         if (as > _sample_peak[j]) {
 166                                 _sample_peak[j] = as;
 167                                 _sample_peak_frame[j] = _done + i;
 168                         }
 169
 170                         if (((_done + i) % _samples_per_point) == 0) {
 171                                 _current[j][AudioPoint::RMS] = sqrt (_current[j][AudioPoint::RMS] / _samples_per_point);
 172                                 _analysis.add_point (j, _current[j]);
 173                                 _current[j] = AudioPoint ();
 174                         }
 175                 }
 176         }
 177
 178         _leqm->add(interleaved);
 179
 180         _done += frames;
 181
 182         DCPTime const length = _playlist->length (_film);
 183         _set_progress ((time.seconds() - _start.seconds()) / (length.seconds() - _start.seconds()));
 184         LOG_DEBUG_AUDIO_ANALYSIS_NC("Frames processed");
 185 }
 186
 187
 188 void
 189 AudioAnalyser::finish ()
 190 {
 191         vector<AudioAnalysis::PeakTime> sample_peak;
 192         for (int i = 0; i < _film->audio_channels(); ++i) {
 193                 sample_peak.push_back (
 194                         AudioAnalysis::PeakTime (_sample_peak[i], DCPTime::from_frames (_sample_peak_frame[i], _film->audio_frame_rate ()))
 195                         );
 196         }
 197         _analysis.set_sample_peak (sample_peak);
 198
 199 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
 200         if (Config::instance()->analyse_ebur128 ()) {
 201                 void* eb = _ebur128->get("Parsed_ebur128_0")->priv;
 202                 vector<float> true_peak;
 203                 for (int i = 0; i < _film->audio_channels(); ++i) {
 204                         true_peak.push_back (av_ebur128_get_true_peaks(eb)[i]);
 205                 }
 206                 _analysis.set_true_peak (true_peak);
 207                 _analysis.set_integrated_loudness (av_ebur128_get_integrated_loudness(eb));
 208                 _analysis.set_loudness_range (av_ebur128_get_loudness_range(eb));
 209         }
 210 #endif
 211
 212         if (_playlist->content().size() == 1) {
 213                 /* If there was only one piece of content in this analysis we may later need to know what its
 214                    gain was when we analysed it.
 215                 */
 216                 if (auto ac = _playlist->content().front()->audio) {
 217                         _analysis.set_analysis_gain (ac->gain());
 218                 }
 219         }
 220
 221         _analysis.set_samples_per_point (_samples_per_point);
 222         _analysis.set_sample_rate (_film->audio_frame_rate ());
 223         _analysis.set_leqm (_leqm->leq_m());
 224 }