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 <dcp/warnings.h>
  34 extern "C" {
  35 #include <leqm_nrt.h>
  36 LIBDCP_DISABLE_WARNINGS
  37 #include <libavutil/channel_layout.h>
  38 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  39 #include <libavfilter/f_ebur128.h>
  40 #endif
  41 LIBDCP_ENABLE_WARNINGS
  42 }
  43
  44
  45 using std::make_shared;
  46 using std::max;
  47 using std::shared_ptr;
  48 using std::vector;
  49 using namespace dcpomatic;
  50
  51
  52 static auto constexpr num_points = 1024;
  53
  54
  55 AudioAnalyser::AudioAnalyser (shared_ptr<const Film> film, shared_ptr<const Playlist> playlist, bool from_zero, std::function<void (float)> set_progress)
  56         : _film (film)
  57         , _playlist (playlist)
  58         , _set_progress (set_progress)
  59 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  60         , _ebur128 (new AudioFilterGraph(film->audio_frame_rate(), film->audio_channels()))
  61 #endif
  62         , _sample_peak (film->audio_channels())
  63         , _sample_peak_frame (film->audio_channels())
  64         , _analysis (film->audio_channels())
  65 {
  66
  67 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  68         _filters.push_back (new Filter("ebur128", "ebur128", "audio", "ebur128=peak=true"));
  69         _ebur128->setup (_filters);
  70 #endif
  71
  72         _current = std::vector<AudioPoint>(_film->audio_channels());
  73
  74         if (!from_zero) {
  75                 _start = _playlist->start().get_value_or(DCPTime());
  76         }
  77
  78         for (int i = 0; i < film->audio_channels(); ++i) {
  79                 _sample_peak[i] = 0;
  80                 _sample_peak_frame[i] = 0;
  81         }
  82
  83         auto add_if_required = [](vector<double>& v, size_t i, double db) {
  84                 if (v.size() > i) {
  85                         v[i] = pow(10, db / 20);
  86                 }
  87         };
  88
  89         int leqm_channels = film->audio_channels();
  90         auto content = _playlist->content();
  91         if (content.size() == 1 && content[0]->audio) {
  92                 leqm_channels = content[0]->audio->mapping().mapped_output_channels().size();
  93         }
  94
  95         /* XXX: is this right?  Especially for more than 5.1? */
  96         vector<double> channel_corrections(leqm_channels, 1);
  97         add_if_required (channel_corrections,  4,   -3); // Ls
  98         add_if_required (channel_corrections,  5,   -3); // Rs
  99         add_if_required (channel_corrections,  6, -144); // HI
 100         add_if_required (channel_corrections,  7, -144); // VI
 101         add_if_required (channel_corrections,  8,   -3); // Lc
 102         add_if_required (channel_corrections,  9,   -3); // Rc
 103         add_if_required (channel_corrections, 10,   -3); // Lc
 104         add_if_required (channel_corrections, 11,   -3); // Rc
 105         add_if_required (channel_corrections, 12, -144); // DBox
 106         add_if_required (channel_corrections, 13, -144); // Sync
 107         add_if_required (channel_corrections, 14, -144); // Sign Language
 108         add_if_required (channel_corrections, 15, -144); // Unused
 109
 110         _leqm.reset(new leqm_nrt::Calculator(
 111                 leqm_channels,
 112                 film->audio_frame_rate(),
 113                 24,
 114                 channel_corrections,
 115                 850, // suggested by leqm_nrt CLI source
 116                 64,  // suggested by leqm_nrt CLI source
 117                 boost::thread::hardware_concurrency()
 118                 ));
 119
 120         DCPTime const length = _playlist->length (_film);
 121
 122         Frame const len = DCPTime (length - _start).frames_round (film->audio_frame_rate());
 123         _samples_per_point = max (int64_t (1), len / num_points);
 124 }
 125
 126
 127 AudioAnalyser::~AudioAnalyser ()
 128 {
 129         for (auto i: _filters) {
 130                 delete const_cast<Filter*> (i);
 131         }
 132 }
 133
 134
 135 void
 136 AudioAnalyser::analyse (shared_ptr<AudioBuffers> b, DCPTime time)
 137 {
 138         LOG_DEBUG_AUDIO_ANALYSIS("AudioAnalyser received %1 frames at %2", b->frames(), to_string(time));
 139         DCPOMATIC_ASSERT (time >= _start);
 140         /* In bug #2364 we had a lot of frames arriving here (~47s worth) which
 141          * caused an OOM error on Windows.  Check for the number of frames being
 142          * reasonable here to make sure we catch this if it happens again.
 143          */
 144         DCPOMATIC_ASSERT(b->frames() < 480000);
 145
 146 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
 147         if (Config::instance()->analyse_ebur128 ()) {
 148                 _ebur128->process (b);
 149         }
 150 #endif
 151
 152         int const frames = b->frames ();
 153         int const channels = b->channels ();
 154         vector<double> interleaved(frames * channels);
 155
 156         for (int j = 0; j < channels; ++j) {
 157                 float const* data = b->data(j);
 158                 for (int i = 0; i < frames; ++i) {
 159                         float s = data[i];
 160
 161                         interleaved[i * channels + j] = s;
 162
 163                         float as = fabsf (s);
 164                         if (as < 10e-7) {
 165                                 /* We may struggle to serialise and recover inf or -inf, so prevent such
 166                                    values by replacing with this (140dB down) */
 167                                 s = as = 10e-7;
 168                         }
 169                         _current[j][AudioPoint::RMS] += pow (s, 2);
 170                         _current[j][AudioPoint::PEAK] = max (_current[j][AudioPoint::PEAK], as);
 171
 172                         if (as > _sample_peak[j]) {
 173                                 _sample_peak[j] = as;
 174                                 _sample_peak_frame[j] = _done + i;
 175                         }
 176
 177                         if (((_done + i) % _samples_per_point) == 0) {
 178                                 _current[j][AudioPoint::RMS] = sqrt (_current[j][AudioPoint::RMS] / _samples_per_point);
 179                                 _analysis.add_point (j, _current[j]);
 180                                 _current[j] = AudioPoint ();
 181                         }
 182                 }
 183         }
 184
 185         _leqm->add(interleaved);
 186
 187         _done += frames;
 188
 189         DCPTime const length = _playlist->length (_film);
 190         _set_progress ((time.seconds() - _start.seconds()) / (length.seconds() - _start.seconds()));
 191         LOG_DEBUG_AUDIO_ANALYSIS_NC("Frames processed");
 192 }
 193
 194
 195 void
 196 AudioAnalyser::finish ()
 197 {
 198         vector<AudioAnalysis::PeakTime> sample_peak;
 199         for (int i = 0; i < _film->audio_channels(); ++i) {
 200                 sample_peak.push_back (
 201                         AudioAnalysis::PeakTime (_sample_peak[i], DCPTime::from_frames (_sample_peak_frame[i], _film->audio_frame_rate ()))
 202                         );
 203         }
 204         _analysis.set_sample_peak (sample_peak);
 205
 206 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
 207         if (Config::instance()->analyse_ebur128 ()) {
 208                 void* eb = _ebur128->get("Parsed_ebur128_0")->priv;
 209                 vector<float> true_peak;
 210                 for (int i = 0; i < _film->audio_channels(); ++i) {
 211                         true_peak.push_back (av_ebur128_get_true_peaks(eb)[i]);
 212                 }
 213                 _analysis.set_true_peak (true_peak);
 214                 _analysis.set_integrated_loudness (av_ebur128_get_integrated_loudness(eb));
 215                 _analysis.set_loudness_range (av_ebur128_get_loudness_range(eb));
 216         }
 217 #endif
 218
 219         if (_playlist->content().size() == 1) {
 220                 /* If there was only one piece of content in this analysis we may later need to know what its
 221                    gain was when we analysed it.
 222                 */
 223                 if (auto ac = _playlist->content().front()->audio) {
 224                         _analysis.set_analysis_gain (ac->gain());
 225                 }
 226         }
 227
 228         _analysis.set_samples_per_point (_samples_per_point);
 229         _analysis.set_sample_rate (_film->audio_frame_rate ());
 230         _analysis.set_leqm (_leqm->leq_m());
 231 }