Use CEF pts for audio

[aiden-jeffrey]

In this change we use the CEF pts (which is in unix epoch ms) to seed our own internal frame time counter. If we identify a drift greater than 1ms (which we need to allow because of integer rounding), then we reset our counter to the CEF pts.

Note I have had to add some latency to the audio "cefGstOffset" (the value we add to the CEF pts to get to gst-land) in order for there to not to be underflow on the audio branch of the pipeline. The value here (4*audio buffer duration) was arrived at by trial and error.

[MathieuDuponchelle]

Nice :) I think the final solution might prove to be slightly more involved, have you tried to determine if there is indeed drift between the system clock that is used by default and the CEF clock?

I think https://gitlab.freedesktop.org/gstreamer/gstreamer/-/blob/main/subprojects/gst-plugins-bad/sys/aja/gstajasrc.cpp?ref_type=heads#L2751 and below could be useful as a reference, @sdroege can you perhaps offer guidance here? :)

[aiden-jeffrey]

I think https://gitlab.freedesktop.org/gstreamer/gstreamer/-/blob/main/subprojects/gst-plugins-bad/sys/aja/gstajasrc.cpp?ref_type=heads#L2751 and below could be useful as a reference, @sdroege can you perhaps offer guidance here? :)

Since this improves things, can we merge as-is?

[MathieuDuponchelle]

Since this improves things, can we merge as-is?

I'm afraid I'd rather hold for now until we've had time to properly review. As discussed in matrix you shouldn't be worried about conflicts, you can consider you have an exclusive lock over the codebase until then :)

[aiden-jeffrey]

Closing in favour of #107