[asterisk-dev] Jitter Buffer

[Steve Kann]

On Oct 23, 2006, at 11:25 AM, John Lange wrote:

> Have a look at the other thread on this topic. This new JB
> implementation seems thoroughly confusing to me at this point.
>
>> From what I understand, if you are doing SIP <-> IAX and you want  
>> the JB
> on the IAX leg you have to enable it in sip.conf(?!).
>
> In my case I need the JB on the SIP leg of a SIP<->ZAP call so I  
> have to
> enable it in zapata.conf.
>
> Can someone bring some clarity to the issue?
>
> BTW, there is a bug in the JB which causes no audio to flow in one
> direction. I'm trying to track it down and will file a bug report if I
> can isolate it.

I've seen this occasionally with the IAX2 implementation.  (note, the  
core jitterbuffer algorithm in use is the same if you're using the  
current IAX2 jitterbuffer integration, or the "adaptive" SIP  
integration).

I'm pretty sure the problem has to do with handling negative offsets  
properly somewhere.  When I wrote this (it seems like it must be  
almost 2 years ago now?),  I probably wasn't very careful around how  
integers wrap and so forth.  I've seen some traces, where when the  
initial timestamps come through with negative delays, the code that  
would normally come and "resync" the jitterbuffer doesn't actually  
run, and then it may take a very long time (meaning, perhaps hours),  
before the adaptation will catch up.

-SteveK