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