Split audio analysis code off from the job.
[dcpomatic.git] / src / lib / audio_analyser.cc
diff --git a/src/lib/audio_analyser.cc b/src/lib/audio_analyser.cc
new file mode 100644 (file)
index 0000000..3caa997
--- /dev/null
@@ -0,0 +1,221 @@
+/*
+    Copyright (C) 2021 Carl Hetherington <cth@carlh.net>
+
+    This file is part of DCP-o-matic.
+
+    DCP-o-matic is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    DCP-o-matic is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with DCP-o-matic.  If not, see <http://www.gnu.org/licenses/>.
+
+*/
+
+
+#include "audio_analyser.h"
+#include "audio_analysis.h"
+#include "audio_buffers.h"
+#include "audio_content.h"
+#include "audio_filter_graph.h"
+#include "audio_point.h"
+#include "config.h"
+#include "dcpomatic_log.h"
+#include "film.h"
+#include "filter.h"
+#include "playlist.h"
+#include "types.h"
+extern "C" {
+#include <leqm_nrt.h>
+#include <libavutil/channel_layout.h>
+#ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
+#include <libavfilter/f_ebur128.h>
+#endif
+}
+
+
+using std::make_shared;
+using std::max;
+using std::shared_ptr;
+using std::vector;
+using namespace dcpomatic;
+
+
+static auto constexpr num_points = 1024;
+
+
+AudioAnalyser::AudioAnalyser (shared_ptr<const Film> film, shared_ptr<const Playlist> playlist, bool from_zero, std::function<void (float)> set_progress)
+       : _film (film)
+       , _playlist (playlist)
+       , _set_progress (set_progress)
+#ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
+       , _ebur128 (new AudioFilterGraph(film->audio_frame_rate(), film->audio_channels()))
+#endif
+       , _sample_peak (new float[film->audio_channels()])
+       , _sample_peak_frame (new Frame[film->audio_channels()])
+       , _analysis (film->audio_channels())
+{
+
+#ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
+       _filters.push_back (new Filter("ebur128", "ebur128", "audio", "ebur128=peak=true"));
+       _ebur128->setup (_filters);
+#endif
+
+       _current = new AudioPoint[_film->audio_channels()];
+
+       if (!from_zero) {
+               _start = _playlist->start().get_value_or(DCPTime());
+       }
+
+       for (int i = 0; i < film->audio_channels(); ++i) {
+               _sample_peak[i] = 0;
+               _sample_peak_frame[i] = 0;
+       }
+
+       auto add_if_required = [](vector<double>& v, size_t i, double db) {
+               if (v.size() > i) {
+                       v[i] = pow(10, db / 20);
+               }
+       };
+
+       /* XXX: is this right?  Especially for more than 5.1? */
+       vector<double> channel_corrections(film->audio_channels(), 1);
+       add_if_required (channel_corrections,  4,   -3); // Ls
+       add_if_required (channel_corrections,  5,   -3); // Rs
+       add_if_required (channel_corrections,  6, -144); // HI
+       add_if_required (channel_corrections,  7, -144); // VI
+       add_if_required (channel_corrections,  8,   -3); // Lc
+       add_if_required (channel_corrections,  9,   -3); // Rc
+       add_if_required (channel_corrections, 10,   -3); // Lc
+       add_if_required (channel_corrections, 11,   -3); // Rc
+       add_if_required (channel_corrections, 12, -144); // DBox
+       add_if_required (channel_corrections, 13, -144); // Sync
+       add_if_required (channel_corrections, 14, -144); // Sign Language
+       add_if_required (channel_corrections, 15, -144); // Unused
+
+       _leqm.reset(new leqm_nrt::Calculator(
+               film->audio_channels(),
+               film->audio_frame_rate(),
+               24,
+               channel_corrections,
+               850, // suggested by leqm_nrt CLI source
+               64,  // suggested by leqm_nrt CLI source
+               boost::thread::hardware_concurrency()
+               ));
+
+       DCPTime const length = _playlist->length (_film);
+
+       Frame const len = DCPTime (length - _start).frames_round (film->audio_frame_rate());
+       _samples_per_point = max (int64_t (1), len / num_points);
+}
+
+
+AudioAnalyser::~AudioAnalyser ()
+{
+       delete[] _current;
+       for (auto i: _filters) {
+               delete const_cast<Filter*> (i);
+       }
+       delete[] _sample_peak;
+       delete[] _sample_peak_frame;
+}
+
+
+void
+AudioAnalyser::analyse (shared_ptr<const AudioBuffers> b, DCPTime time)
+{
+       LOG_DEBUG_AUDIO_ANALYSIS("Received %1 frames at %2", b->frames(), to_string(time));
+       DCPOMATIC_ASSERT (time >= _start);
+
+#ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
+       if (Config::instance()->analyse_ebur128 ()) {
+               _ebur128->process (b);
+       }
+#endif
+
+       int const frames = b->frames ();
+       int const channels = b->channels ();
+       vector<double> interleaved(frames * channels);
+
+       for (int j = 0; j < channels; ++j) {
+               float* data = b->data(j);
+               for (int i = 0; i < frames; ++i) {
+                       float s = data[i];
+
+                       interleaved[i * channels + j] = s;
+
+                       float as = fabsf (s);
+                       if (as < 10e-7) {
+                               /* We may struggle to serialise and recover inf or -inf, so prevent such
+                                  values by replacing with this (140dB down) */
+                               s = as = 10e-7;
+                       }
+                       _current[j][AudioPoint::RMS] += pow (s, 2);
+                       _current[j][AudioPoint::PEAK] = max (_current[j][AudioPoint::PEAK], as);
+
+                       if (as > _sample_peak[j]) {
+                               _sample_peak[j] = as;
+                               _sample_peak_frame[j] = _done + i;
+                       }
+
+                       if (((_done + i) % _samples_per_point) == 0) {
+                               _current[j][AudioPoint::RMS] = sqrt (_current[j][AudioPoint::RMS] / _samples_per_point);
+                               _analysis.add_point (j, _current[j]);
+                               _current[j] = AudioPoint ();
+                       }
+               }
+       }
+
+       _leqm->add(interleaved);
+
+       _done += frames;
+
+       DCPTime const length = _playlist->length (_film);
+       _set_progress ((time.seconds() - _start.seconds()) / (length.seconds() - _start.seconds()));
+       LOG_DEBUG_AUDIO_ANALYSIS_NC("Frames processed");
+}
+
+
+void
+AudioAnalyser::finish ()
+{
+       vector<AudioAnalysis::PeakTime> sample_peak;
+       for (int i = 0; i < _film->audio_channels(); ++i) {
+               sample_peak.push_back (
+                       AudioAnalysis::PeakTime (_sample_peak[i], DCPTime::from_frames (_sample_peak_frame[i], _film->audio_frame_rate ()))
+                       );
+       }
+       _analysis.set_sample_peak (sample_peak);
+
+#ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
+       if (Config::instance()->analyse_ebur128 ()) {
+               void* eb = _ebur128->get("Parsed_ebur128_0")->priv;
+               vector<float> true_peak;
+               for (int i = 0; i < _film->audio_channels(); ++i) {
+                       true_peak.push_back (av_ebur128_get_true_peaks(eb)[i]);
+               }
+               _analysis.set_true_peak (true_peak);
+               _analysis.set_integrated_loudness (av_ebur128_get_integrated_loudness(eb));
+               _analysis.set_loudness_range (av_ebur128_get_loudness_range(eb));
+       }
+#endif
+
+       if (_playlist->content().size() == 1) {
+               /* If there was only one piece of content in this analysis we may later need to know what its
+                  gain was when we analysed it.
+               */
+               if (auto ac = _playlist->content().front()->audio) {
+                       _analysis.set_analysis_gain (ac->gain());
+               }
+       }
+
+       _analysis.set_samples_per_point (_samples_per_point);
+       _analysis.set_sample_rate (_film->audio_frame_rate ());
+       _analysis.set_leqm (_leqm->leq_m());
+}