We are experiencing audio sync issues during playback on fMP4 HLS live streams (HLS and LL-HLS) on Apple devices only (iOS and macOS) and we're not sure what's causing the problem. The issue does not occur during playback on Windows or Android platforms.
During playback in Safari, everything is fine until the sync gets lost suddenly, usually 5-10 minutes after playback begins. The extent of the desync varies but is very noticeable when it does - usually in the 15-30 frame range. Sync is always restored when restarting the player, until it becomes lost again some minutes later.
We are capturing the streams on iPhone devices and encoding HEVC / AAC-LC at 30fps locally on the device, and then sending to a media server for further processing. We then transcode the source stream and create multiple variations at different bitrates (HEVC). Because we are streaming from mobile devices in the field, during our server-side transcoding we set a constant 30fps frame rate in case of drops due to network issues. I should add that the issue occurs just as much with h264 as HEVC (we've tested many different combinations of input/output formats and protocols).
Regardless of whether we playback the source stream, the individual transcoded variations, or the ABR playlist with all variations, the sync problem appears in the same manner.
One interesting note. The issue seldom occurs on one of our older devices, an iPhone 6s Plus running a slightly older iOS version (14.4.1).
We suspect it has something to do with discontinuities inherent in our input streams that are not being corrected during our normalization/transcoding process. The Apple player is not compensating as other players are doing on other platforms.
We've run the Apple MediaStreamValidator validator tool and discovered multiple "must fix" issues - but it's not clear which of these, if any, are causing our problems. See output attached.
Also, here is the full HLS report from the validator tool (in PNG format due to file restrictions here):
Happy to share more details or run more tests. We've been trying to debug this for weeks now. Thanks for your help.
