- if (content->content_audio_frame_rate() != content->output_audio_frame_rate() && content->audio_channels ()) {
- _resampler.reset (
- new Resampler (
- content->content_audio_frame_rate(),
- content->output_audio_frame_rate(),
- content->audio_channels()
- )
+ 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) {
+ 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;
+ 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))