X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Faudio_decoder.cc;h=61ff5d265c526e6de31e44d5bfb7eb6972883fba;hb=2b0e9dd97a5773f52eba5704903b82e90f4c6f63;hp=f7f147bd92e9b80f58cc46f5bd91c34fb551d6fd;hpb=0241df1707c7ea5658f471828ff6dc944e21af42;p=dcpomatic.git diff --git a/src/lib/audio_decoder.cc b/src/lib/audio_decoder.cc index f7f147bd9..61ff5d265 100644 --- a/src/lib/audio_decoder.cc +++ b/src/lib/audio_decoder.cc @@ -18,6 +18,7 @@ */ + #include "audio_decoder.h" #include "audio_buffers.h" #include "audio_content.h" @@ -29,14 +30,14 @@ #include "i18n.h" + using std::cout; -using std::map; -using std::pair; using std::shared_ptr; using std::make_shared; using boost::optional; using namespace dcpomatic; + AudioDecoder::AudioDecoder (Decoder* parent, shared_ptr content, bool fast) : DecoderPart (parent) , _content (content) @@ -48,41 +49,39 @@ AudioDecoder::AudioDecoder (Decoder* parent, shared_ptr cont } } + /** @param time_already_delayed true if the delay should not be added to time */ void -AudioDecoder::emit (shared_ptr film, AudioStreamPtr stream, shared_ptr data, ContentTime time, bool time_already_delayed) +AudioDecoder::emit(shared_ptr film, AudioStreamPtr stream, shared_ptr data, ContentTime time, bool flushing) { if (ignore ()) { return; } - /* Amount of error we will tolerate on audio timestamps; see comment below. - * We'll use 1 24fps video frame at 48kHz as this seems to be roughly how - * ffplay does it. - */ - static Frame const slack_frames = 48000 / 24; - int const resampled_rate = _content->resampled_frame_rate(film); - if (!time_already_delayed) { + if (!flushing) { time += ContentTime::from_seconds (_content->delay() / 1000.0); } - auto reset = false; - 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 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. - */ - if (_content->delay() > 0) { - /* Insert silence to give the delay */ - silence (_content->delay ()); - } - reset = true; - } else if (std::abs(_positions[stream] - time.frames_round(resampled_rate)) > slack_frames) { - reset = true; + /* 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], @@ -91,13 +90,17 @@ AudioDecoder::emit (shared_ptr film, AudioStreamPtr stream, shared_p ); } - if (reset) { + 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; auto i = _resamplers.find(stream); - if (i != _resamplers.end ()) { + if (i != _resamplers.end()) { resampler = i->second; } else { if (stream->frame_rate() != resampled_rate) { @@ -116,18 +119,31 @@ AudioDecoder::emit (shared_ptr film, AudioStreamPtr stream, shared_p } } - if (resampler) { - auto 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 (shared_ptr film, AudioStreamPtr stream) const @@ -137,6 +153,7 @@ AudioDecoder::stream_position (shared_ptr film, AudioStreamPtr strea return ContentTime::from_frames (i->second, _content->resampled_frame_rate(film)); } + boost::optional AudioDecoder::position (shared_ptr film) const { @@ -151,6 +168,7 @@ AudioDecoder::position (shared_ptr film) const return p; } + void AudioDecoder::seek () { @@ -164,6 +182,7 @@ AudioDecoder::seek () } } + void AudioDecoder::flush () { @@ -177,17 +196,18 @@ AudioDecoder::flush () 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) { - for (auto i: _content->streams()) { - int const samples = ContentTime::from_seconds(milliseconds / 1000.0).frames_round(i->frame_rate()); - auto silence = make_shared(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(stream->channels(), samples); + silence->make_silent (); + Data (stream, ContentAudio(silence, _positions[stream])); }