Gstreamer HLS Help Request

[David Manpearl]

I am trying to create an HLS streaming solution that generates arbitrary
transport stream (.ts) chunks that will fit together and play properly
inside mobile video players without using the .m3u8 DISCONTINUITY flag, but
I am having difficulty offsetting the initial pts values of the first audio
and video frames of each chunk relative to each other.

I am currently using "mpegtsmux" and have custom software to write the
manifest files. I would like to modify the pts values by calling functions
on the pipeline or its elements, but if this is impossible I am prepared to
write a Gstreamer plugin, possibly based on mpegtsmux and with some of the
features of hlssink and/or hlssink2.

Goals:
1. How to offset the presentation time (pts) of the first video frame in a
chunk relative to the first audio frame.
2. How to stop encoding after an exact number of audio packets have been
written (i.e. 93 or 94 for 2-second chunks at 30fps with 48KHz audio).
3. How to determine if this can be done using existing plugins via
modification of the parameters sent to the various GstElements instead of
having to write a new plugin.
4. (optional) Understand how and why hlssink starts the first video frame
of the first chunk in a sequence with pts value 3600.

It appears that the solution might use "gst_structure_get_clock_time()" for
mpegtsmux or other video and/or audio GstElements. But, I am confused in
the hlssink solution which reads the "running_time" into
"current_running_time_start" at the beginning of each chunk and also writes
the difference (running_time - current_running_time_start) into the .m3u8
Manifest file; but I cannot see or understand where it is getting written
into the Pipeline for generation of the .ts media file component.

I am hoping/requesting any advice on how create a stand-alone .ts chunk in
which the first video frame starts after the initial audio packet as per
the "Goals" above. If anyone has pointers I would greatly appreciate your
help.

Backstory: The goal for our users is to allow them to seek to an arbitrary
location in a video that has not yet been completely transcoded, and have
the server seek to the new location and re-start transcoding from there in
such a way that we can come back at a later time and fill in the missing
chunks. hlssink/hlssink2, OTOH, transcodes an entire video file into
multiple chunks from start to finish in a single pass.

Sincerely and thank you
 - David

P.S. Prior to joining gstreamer-devel, I also posted a version of this
question here:
https://stackoverflow.com/questions/50789447/gstreamer-elementary-transport-stream-ts-chunk-audio-video-offset
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/gstreamer-devel/attachments/20180612/576552bf/attachment.html>