How much earlier than its audio a frame is shown, to make up for the video path's longer encode and jitter buffer in the room. 0 until a live run measures it.
The epoch a sample belongs to, from its track and stream time: the current generation, and a new segment whenever the track's time goes backwards.
The sample's track.
Its presentation time (s).
Whether a piece of this epoch is stale (decoded before the last reset).
A new init segment: what comes next is a new stream, whose times may start over. Both tracks move to a new segment; what was decoded before still plays (a turn's tail keeps its frames while the next turn's stream begins).
Records that a piece of the avatar's voice entered the voice queue: its stream time and length, and how much the queue had taken before it. Stale pieces are ignored (the caller does not queue them).
The piece's epoch.
Its stream time (s).
Its length (s).
The queue's total before it (ms).
A barge-in: everything decoded before is stale, and the timelines start over.
When a frame is due.
The frame's epoch.
Its presentation time (s).
How long it has waited since it was decoded (ms).
Places an avatar's frames against its voice's playout. One per avatar stream.