Incorrect timestamps from MTL when receiving audio (st30p)

[cwhite102]

We noticed a lip sync issue and spent time narrowing it down to the MTL Audio Rx pipeline.
It appears as though the timestamps provided by MTL for received audio frames are not based on the first packet's RTP timestamp (as they should be), but instead, on the last packet received for the frame of data.

In pipeline mode, we typically configure MTL to provide 10ms worth of audio data at a time. Using an A/V offset test signal input, we are observing the received audio being offset from the video by 9ms. If we change the settings to receive 20ms of data, the A/V offset becomes 19ms, etc..

The timestamp reported with the frame in pipeline mode should correspond to the RTP timestamp of the beginning of the data, not the final 2110 packet.

(MTL main branch)

[cwhite102]

The offending code appears to be

st_rx_audio_session.c
rx_audio_session_handle_frame_pkt()

Where it writes meta->timestamp and meta->rtp_timestamp (not sure why there's two, they get the same value)
(at or around line 359)

Any fix should also consider out-of-order packet arrival if that is a concern at this point in the packet processing.