- bool 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) {