src/lib/analyse_audio_job.cc

   1 /*
   2     Copyright (C) 2012-2018 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 #include "audio_analysis.h"
  22 #include "audio_buffers.h"
  23 #include "analyse_audio_job.h"
  24 #include "audio_content.h"
  25 #include "compose.hpp"
  26 #include "dcpomatic_log.h"
  27 #include "film.h"
  28 #include "player.h"
  29 #include "playlist.h"
  30 #include "filter.h"
  31 #include "audio_filter_graph.h"
  32 #include "config.h"
  33 extern "C" {
  34 #include <leqm_nrt.h>
  35 #include <libavutil/channel_layout.h>
  36 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  37 #include <libavfilter/f_ebur128.h>
  38 #endif
  39 }
  40 #include <iostream>
  41
  42 #include "i18n.h"
  43
  44 using std::string;
  45 using std::vector;
  46 using std::max;
  47 using std::min;
  48 using std::cout;
  49 using std::shared_ptr;
  50 using std::dynamic_pointer_cast;
  51 using namespace dcpomatic;
  52 #if BOOST_VERSION >= 106100
  53 using namespace boost::placeholders;
  54 #endif
  55
  56 int const AnalyseAudioJob::_num_points = 1024;
  57
  58 static void add_if_required(vector<double>& v, size_t i, double db)
  59 {
  60         if (v.size() > i) {
  61                 v[i] = pow(10, db / 20);
  62         }
  63 }
  64
  65 /** @param from_zero true to analyse audio from time 0 in the playlist, otherwise begin at Playlist::start */
  66 AnalyseAudioJob::AnalyseAudioJob (shared_ptr<const Film> film, shared_ptr<const Playlist> playlist, bool from_zero)
  67         : Job (film)
  68         , _playlist (playlist)
  69         , _path (film->audio_analysis_path(playlist))
  70         , _from_zero (from_zero)
  71         , _done (0)
  72         , _samples_per_point (1)
  73         , _current (0)
  74         , _sample_peak (new float[film->audio_channels()])
  75         , _sample_peak_frame (new Frame[film->audio_channels()])
  76 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  77         , _ebur128 (new AudioFilterGraph (film->audio_frame_rate(), film->audio_channels()))
  78 #endif
  79 {
  80         LOG_DEBUG_AUDIO_ANALYSIS_NC("AnalyseAudioJob::AnalyseAudioJob");
  81
  82 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  83         _filters.push_back (new Filter ("ebur128", "ebur128", "audio", "ebur128=peak=true"));
  84         _ebur128->setup (_filters);
  85 #endif
  86
  87         for (int i = 0; i < film->audio_channels(); ++i) {
  88                 _sample_peak[i] = 0;
  89                 _sample_peak_frame[i] = 0;
  90         }
  91
  92         if (!_from_zero) {
  93                 _start = _playlist->start().get_value_or(DCPTime());
  94         }
  95
  96         /* XXX: is this right?  Especially for more than 5.1? */
  97         vector<double> channel_corrections(film->audio_channels(), 1);
  98         add_if_required (channel_corrections,  4,   -3); // Ls
  99         add_if_required (channel_corrections,  5,   -3); // Rs
 100         add_if_required (channel_corrections,  6, -144); // HI
 101         add_if_required (channel_corrections,  7, -144); // VI
 102         add_if_required (channel_corrections,  8,   -3); // Lc
 103         add_if_required (channel_corrections,  9,   -3); // Rc
 104         add_if_required (channel_corrections, 10,   -3); // Lc
 105         add_if_required (channel_corrections, 11,   -3); // Rc
 106         add_if_required (channel_corrections, 12, -144); // DBox
 107         add_if_required (channel_corrections, 13, -144); // Sync
 108         add_if_required (channel_corrections, 14, -144); // Sign Language
 109         add_if_required (channel_corrections, 15, -144); // Unused
 110
 111         _leqm.reset(new leqm_nrt::Calculator(
 112                 film->audio_channels(),
 113                 film->audio_frame_rate(),
 114                 24,
 115                 channel_corrections,
 116                 850, // suggested by leqm_nrt CLI source
 117                 64,  // suggested by leqm_nrt CLI source
 118                 boost::thread::hardware_concurrency()
 119                 ));
 120 }
 121
 122 AnalyseAudioJob::~AnalyseAudioJob ()
 123 {
 124         stop_thread ();
 125         for (auto i: _filters) {
 126                 delete const_cast<Filter*> (i);
 127         }
 128         delete[] _current;
 129         delete[] _sample_peak;
 130         delete[] _sample_peak_frame;
 131 }
 132
 133 string
 134 AnalyseAudioJob::name () const
 135 {
 136         return _("Analysing audio");
 137 }
 138
 139 string
 140 AnalyseAudioJob::json_name () const
 141 {
 142         return N_("analyse_audio");
 143 }
 144
 145 void
 146 AnalyseAudioJob::run ()
 147 {
 148         LOG_DEBUG_AUDIO_ANALYSIS_NC("AnalyseAudioJob::run");
 149
 150         shared_ptr<Player> player (new Player(_film, _playlist));
 151         player->set_ignore_video ();
 152         player->set_ignore_text ();
 153         player->set_fast ();
 154         player->set_play_referenced ();
 155         player->Audio.connect (bind (&AnalyseAudioJob::analyse, this, _1, _2));
 156
 157         DCPTime const length = _playlist->length (_film);
 158
 159         Frame const len = DCPTime (length - _start).frames_round (_film->audio_frame_rate());
 160         _samples_per_point = max (int64_t (1), len / _num_points);
 161
 162         delete[] _current;
 163         _current = new AudioPoint[_film->audio_channels ()];
 164         _analysis.reset (new AudioAnalysis (_film->audio_channels ()));
 165
 166         bool has_any_audio = false;
 167         for (auto c: _playlist->content()) {
 168                 if (c->audio) {
 169                         has_any_audio = true;
 170                 }
 171         }
 172
 173         if (has_any_audio) {
 174                 LOG_DEBUG_AUDIO_ANALYSIS("Seeking to %1", to_string(_start));
 175                 player->seek (_start, true);
 176                 _done = 0;
 177                 LOG_DEBUG_AUDIO_ANALYSIS("Starting loop for playlist of length %1", to_string(length));
 178                 while (!player->pass ()) {}
 179         }
 180
 181         LOG_DEBUG_AUDIO_ANALYSIS_NC("Loop complete");
 182
 183         vector<AudioAnalysis::PeakTime> sample_peak;
 184         for (int i = 0; i < _film->audio_channels(); ++i) {
 185                 sample_peak.push_back (
 186                         AudioAnalysis::PeakTime (_sample_peak[i], DCPTime::from_frames (_sample_peak_frame[i], _film->audio_frame_rate ()))
 187                         );
 188         }
 189         _analysis->set_sample_peak (sample_peak);
 190
 191 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
 192         if (Config::instance()->analyse_ebur128 ()) {
 193                 void* eb = _ebur128->get("Parsed_ebur128_0")->priv;
 194                 vector<float> true_peak;
 195                 for (int i = 0; i < _film->audio_channels(); ++i) {
 196                         true_peak.push_back (av_ebur128_get_true_peaks(eb)[i]);
 197                 }
 198                 _analysis->set_true_peak (true_peak);
 199                 _analysis->set_integrated_loudness (av_ebur128_get_integrated_loudness(eb));
 200                 _analysis->set_loudness_range (av_ebur128_get_loudness_range(eb));
 201         }
 202 #endif
 203
 204         if (_playlist->content().size() == 1) {
 205                 /* If there was only one piece of content in this analysis we may later need to know what its
 206                    gain was when we analysed it.
 207                 */
 208                 shared_ptr<const AudioContent> ac = _playlist->content().front()->audio;
 209                 if (ac) {
 210                         _analysis->set_analysis_gain (ac->gain());
 211                 }
 212         }
 213
 214         _analysis->set_samples_per_point (_samples_per_point);
 215         _analysis->set_sample_rate (_film->audio_frame_rate ());
 216         _analysis->set_leqm (_leqm->leq_m());
 217         _analysis->write (_path);
 218
 219         LOG_DEBUG_AUDIO_ANALYSIS_NC("Job finished");
 220         set_progress (1);
 221         set_state (FINISHED_OK);
 222 }
 223
 224 void
 225 AnalyseAudioJob::analyse (shared_ptr<const AudioBuffers> b, DCPTime time)
 226 {
 227         LOG_DEBUG_AUDIO_ANALYSIS("Received %1 frames at %2", b->frames(), to_string(time));
 228         DCPOMATIC_ASSERT (time >= _start);
 229
 230 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
 231         if (Config::instance()->analyse_ebur128 ()) {
 232                 _ebur128->process (b);
 233         }
 234 #endif
 235
 236         int const frames = b->frames ();
 237         int const channels = b->channels ();
 238         vector<double> interleaved(frames * channels);
 239
 240         for (int j = 0; j < channels; ++j) {
 241                 float* data = b->data(j);
 242                 for (int i = 0; i < frames; ++i) {
 243                         float s = data[i];
 244
 245                         interleaved[i * channels + j] = s;
 246
 247                         float as = fabsf (s);
 248                         if (as < 10e-7) {
 249                                 /* We may struggle to serialise and recover inf or -inf, so prevent such
 250                                    values by replacing with this (140dB down) */
 251                                 s = as = 10e-7;
 252                         }
 253                         _current[j][AudioPoint::RMS] += pow (s, 2);
 254                         _current[j][AudioPoint::PEAK] = max (_current[j][AudioPoint::PEAK], as);
 255
 256                         if (as > _sample_peak[j]) {
 257                                 _sample_peak[j] = as;
 258                                 _sample_peak_frame[j] = _done + i;
 259                         }
 260
 261                         if (((_done + i) % _samples_per_point) == 0) {
 262                                 _current[j][AudioPoint::RMS] = sqrt (_current[j][AudioPoint::RMS] / _samples_per_point);
 263                                 _analysis->add_point (j, _current[j]);
 264                                 _current[j] = AudioPoint ();
 265                         }
 266                 }
 267         }
 268
 269         _leqm->add(interleaved);
 270
 271         _done += frames;
 272
 273         DCPTime const length = _playlist->length (_film);
 274         set_progress ((time.seconds() - _start.seconds()) / (length.seconds() - _start.seconds()));
 275         LOG_DEBUG_AUDIO_ANALYSIS_NC("Frames processed");
 276 }