[AG-TECH] Audio losses synch (more technical questions)
spb at epcc.ed.ac.uk
Mon Jun 10 04:17:20 CDT 2002
On Fri, 7 Jun 2002, White, Derek [NCSUS Non J&J] wrote:
> I am assuming that the key for lip synch is that rat calculates the
> end-to-end delay between encoding and final playback, and then rat sends
> rtp.source.playout messages to the local vic. NTP is needed to get accurate
> time differences. Vic can then adjust it's playback to match rat.
> This actually works to a point - vic gets the rtp.source.playout message,
> but then can't figure out who the sender is! Vic looks through the
> cname_hash_table, which is empty. Rat never sent an rtp.source.cname message
> to any vic (it does send them to other rat processes though).
> I tried a hack where vic sends an rtp.query message to all rats if it gets
> an unknown rtp.source.playout message, which is a little too late, and rat
> starts getting assertion failures soon after.
> I'm not sure where the breakdown occurred. Should rat broadcast cname
> messages to all vics on occasion? Should vic ask for cname at a better time?
I asked a similar question on the UCL rat users list last week.
rat is supposed to be sending the rtp.source.cnames to vic as well.
(I agree with you it does not)
Apparantly the rat developers are not using/testing with vic.
It should be an easy fix to add this to rat.
My understanding of the source is that the rtp.playout messages are
ignored unless you have already seen a cname message for that session.
At the moment this would not be enough to fix a standard AG node, if audio
and video sources are from different machines the cnames will not match
anyway as these are derived from the IP address.
> ps. This is with rat 4.2.16 built from cvs source. I spent a day trying to
> get vic 2.8.ucl-1.1.3-AG to talk to rat 4.2.21, which is hopeless. The newer
> mbus code does encryption and authentication in the opposite order, and over
> a different part of the message! They silently refuse to talk at all.
I managed to compile up the AG vic code linked against the most up to date
UCL common library. (There is some new crypto code that has been
seperatly added to both development trees but if you delete the files
from vic the program still links ok) This might be enough to fix this
problem. It should be as all the mbus code would be from the same
|epcc| Dr Stephen P Booth Project Manager |epcc|
|epcc| s.booth at epcc.ed.ac.uk Phone 0131 650 5746 |epcc|
More information about the ag-tech