/*
- Copyright (C) 2012-2018 Carl Hetherington <cth@carlh.net>
+ Copyright (C) 2012-2021 Carl Hetherington <cth@carlh.net>
This file is part of DCP-o-matic.
*/
+
#include "audio_decoder.h"
#include "audio_buffers.h"
#include "audio_content.h"
+#include "dcpomatic_log.h"
#include "log.h"
#include "resampler.h"
#include "compose.hpp"
-#include <boost/foreach.hpp>
#include <iostream>
#include "i18n.h"
-#define LOG_GENERAL(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_GENERAL);
using std::cout;
-using std::map;
-using std::pair;
-using boost::shared_ptr;
+using std::shared_ptr;
+using std::make_shared;
using boost::optional;
+using namespace dcpomatic;
+
-AudioDecoder::AudioDecoder (Decoder* parent, shared_ptr<const AudioContent> content, shared_ptr<Log> log, bool fast)
- : DecoderPart (parent, log)
+AudioDecoder::AudioDecoder (Decoder* parent, shared_ptr<const AudioContent> content, bool fast)
+ : DecoderPart (parent)
, _content (content)
, _fast (fast)
{
/* Set up _positions so that we have one for each stream */
- BOOST_FOREACH (AudioStreamPtr i, content->streams ()) {
+ for (auto i: content->streams ()) {
_positions[i] = 0;
}
}
+
+/** @param time_already_delayed true if the delay should not be added to time */
void
-AudioDecoder::emit (AudioStreamPtr stream, shared_ptr<const AudioBuffers> data, ContentTime time)
+AudioDecoder::emit(shared_ptr<const Film> film, AudioStreamPtr stream, shared_ptr<const AudioBuffers> data, ContentTime time, bool flushing)
{
if (ignore ()) {
return;
}
- if (_positions[stream] == 0) {
- /* This is the first data we have received since initialisation or seek. Set
- the position based on the ContentTime that was given. After this first time
- we just count samples, as it seems that ContentTimes are unreliable from
- FFmpegDecoder (not quite continuous; perhaps due to some rounding error).
- */
- if (_content->delay() > 0) {
- /* Insert silence to give the delay */
- silence (_content->delay ());
- }
+ int const resampled_rate = _content->resampled_frame_rate(film);
+ if (!flushing) {
time += ContentTime::from_seconds (_content->delay() / 1000.0);
- _positions[stream] = time.frames_round (_content->resampled_frame_rate ());
+ }
+
+ /* Amount of error we will tolerate on audio timestamps; see comment below.
+ * We'll use 1 24fps video frame as this seems to be roughly how ffplay does it.
+ */
+ Frame const slack_frames = resampled_rate / 24;
+
+ /* first_since_seek is set to true if this is the first data we have
+ received since initialisation or seek. We'll set the position based
+ on the ContentTime that was given. After this first time we just
+ count samples unless the timestamp is more than slack_frames away
+ from where we think it should be. This is because ContentTimes seem
+ to be slightly unreliable from FFmpegDecoder (i.e. not sample
+ accurate), but we still need to obey them sometimes otherwise we get
+ sync problems such as #1833.
+ */
+
+ auto const first_since_seek = _positions[stream] == 0;
+ auto const need_reset = !first_since_seek && (std::abs(_positions[stream] - time.frames_round(resampled_rate)) > slack_frames);
+
+ if (need_reset) {
+ LOG_GENERAL (
+ "Reset audio position: was %1, new data at %2, slack: %3 frames",
+ _positions[stream],
+ time.frames_round(resampled_rate),
+ std::abs(_positions[stream] - time.frames_round(resampled_rate))
+ );
+ }
+
+ if (first_since_seek || need_reset) {
+ _positions[stream] = time.frames_round (resampled_rate);
+ }
+
+ if (first_since_seek && _content->delay() > 0) {
+ silence (stream, _content->delay());
}
shared_ptr<Resampler> resampler;
- ResamplerMap::iterator i = _resamplers.find(stream);
- if (i != _resamplers.end ()) {
+ auto i = _resamplers.find(stream);
+ if (i != _resamplers.end()) {
resampler = i->second;
} else {
- if (stream->frame_rate() != _content->resampled_frame_rate()) {
+ if (stream->frame_rate() != resampled_rate) {
LOG_GENERAL (
"Creating new resampler from %1 to %2 with %3 channels",
stream->frame_rate(),
- _content->resampled_frame_rate(),
+ resampled_rate,
stream->channels()
);
- resampler.reset (new Resampler (stream->frame_rate(), _content->resampled_frame_rate(), stream->channels()));
+ resampler = make_shared<Resampler>(stream->frame_rate(), resampled_rate, stream->channels());
if (_fast) {
resampler->set_fast ();
}
}
}
- if (resampler) {
- shared_ptr<const AudioBuffers> ro = resampler->run (data);
- if (ro->frames() == 0) {
+ if (resampler && !flushing) {
+ /* It can be the the data here has a different number of channels than the stream
+ * it comes from (e.g. the files decoded by FFmpegDecoder sometimes have a random
+ * frame, often at the end, with more channels). Insert silence or discard channels
+ * here.
+ */
+ if (resampler->channels() != data->channels()) {
+ LOG_WARNING("Received audio data with an unexpected channel count of %1 instead of %2", data->channels(), resampler->channels());
+ auto data_copy = data->clone();
+ data_copy->set_channels(resampler->channels());
+ data = resampler->run(data_copy);
+ } else {
+ data = resampler->run(data);
+ }
+
+ if (data->frames() == 0) {
return;
}
- data = ro;
}
Data(stream, ContentAudio (data, _positions[stream]));
_positions[stream] += data->frames();
}
+
/** @return Time just after the last thing that was emitted from a given stream */
ContentTime
-AudioDecoder::stream_position (AudioStreamPtr stream) const
+AudioDecoder::stream_position (shared_ptr<const Film> film, AudioStreamPtr stream) const
{
- PositionMap::const_iterator i = _positions.find (stream);
+ auto i = _positions.find (stream);
DCPOMATIC_ASSERT (i != _positions.end ());
- return ContentTime::from_frames (i->second, _content->resampled_frame_rate());
+ return ContentTime::from_frames (i->second, _content->resampled_frame_rate(film));
}
-ContentTime
-AudioDecoder::position () const
+
+boost::optional<ContentTime>
+AudioDecoder::position (shared_ptr<const Film> film) const
{
optional<ContentTime> p;
- for (PositionMap::const_iterator i = _positions.begin(); i != _positions.end(); ++i) {
- ContentTime const ct = stream_position (i->first);
+ for (auto i: _positions) {
+ auto const ct = stream_position (film, i.first);
if (!p || ct < *p) {
p = ct;
}
}
- return p.get_value_or(ContentTime());
+ return p;
}
+
void
AudioDecoder::seek ()
{
- for (ResamplerMap::iterator i = _resamplers.begin(); i != _resamplers.end(); ++i) {
- i->second->flush ();
- i->second->reset ();
+ for (auto i: _resamplers) {
+ i.second->flush ();
+ i.second->reset ();
}
- for (PositionMap::iterator i = _positions.begin(); i != _positions.end(); ++i) {
- i->second = 0;
+ for (auto& i: _positions) {
+ i.second = 0;
}
}
+
void
AudioDecoder::flush ()
{
- for (ResamplerMap::iterator i = _resamplers.begin(); i != _resamplers.end(); ++i) {
- shared_ptr<const AudioBuffers> ro = i->second->flush ();
+ for (auto const& i: _resamplers) {
+ auto ro = i.second->flush ();
if (ro->frames() > 0) {
- Data (i->first, ContentAudio (ro, _positions[i->first]));
- _positions[i->first] += ro->frames();
+ Data (i.first, ContentAudio (ro, _positions[i.first]));
+ _positions[i.first] += ro->frames();
}
}
if (_content->delay() < 0) {
/* Finish off with the gap caused by the delay */
- silence (-_content->delay ());
+ for (auto stream: _content->streams()) {
+ silence (stream, -_content->delay());
+ }
}
}
+
void
-AudioDecoder::silence (int milliseconds)
+AudioDecoder::silence (AudioStreamPtr stream, int milliseconds)
{
- BOOST_FOREACH (AudioStreamPtr i, _content->streams ()) {
- int const samples = ContentTime::from_seconds(milliseconds / 1000.0).frames_round(i->frame_rate());
- shared_ptr<AudioBuffers> silence (new AudioBuffers (i->channels(), samples));
- silence->make_silent ();
- Data (i, ContentAudio (silence, _positions[i]));
- }
+ int const samples = ContentTime::from_seconds(milliseconds / 1000.0).frames_round(stream->frame_rate());
+ auto silence = make_shared<AudioBuffers>(stream->channels(), samples);
+ silence->make_silent ();
+ Data (stream, ContentAudio(silence, _positions[stream]));
}
projects / dcpomatic.git / blobdiff