X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Faudio_decoder.cc;h=61ff5d265c526e6de31e44d5bfb7eb6972883fba;hb=c7f45cd00b94393f6e15428a2ea256995f890412;hp=2d02043b5948d1071d0de18982e14abfcf03c12e;hpb=2e2f11b29651cffe37c64275dbd45c7563310020;p=dcpomatic.git

diff --git a/src/lib/audio_decoder.cc b/src/lib/audio_decoder.cc
index 2d02043b5..61ff5d265 100644
--- a/src/lib/audio_decoder.cc
+++ b/src/lib/audio_decoder.cc
@@ -52,39 +52,36 @@ AudioDecoder::AudioDecoder (Decoder* parent, shared_ptr<const AudioContent> cont
 
 /** @param time_already_delayed true if the delay should not be added to time */
 void
-AudioDecoder::emit (shared_ptr<const Film> film, AudioStreamPtr stream, shared_ptr<const AudioBuffers> data, ContentTime time, bool time_already_delayed)
+AudioDecoder::emit(shared_ptr<const Film> film, AudioStreamPtr stream, shared_ptr<const AudioBuffers> 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],
@@ -93,10 +90,14 @@ AudioDecoder::emit (shared_ptr<const Film> 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> resampler;
 	auto i = _resamplers.find(stream);
 	if (i != _resamplers.end()) {
@@ -118,12 +119,24 @@ AudioDecoder::emit (shared_ptr<const Film> 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]));
@@ -183,18 +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<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]));
 }