[Asterisk-Dev] Re: [Asterisk-cvs] asterisk channel.c,1.187,1.188

[Kevin P. Fleming]

Steve Kann wrote:

> I've been meaning to chat with you about this.   So now, the only time 
> that we go through SLINEAR when we wouldn't have before is in the alaw 
> <-> ulaw case, since all the other translators have slinear on one end..

That's correct. The only effective result of this change (other than 
what you mention above) is that any application that copies frames from 
the channel (ChanSpy et al.) now get them in SLINEAR when transcoding is 
happening, rather than in the native format of one of the two ends 
(depending on the direction). This saves transcoding resources, since 
they don't have to be handled again on their way into the listening 
application/channel.

> But, forcing transcoding is definitely a bad thing, not just for 
> performance, but also for quality, as you lose quality every time you 
> transcode (and it certainly _is_ noticable).

Absolutely, and I'm really surprised that nobody noticed this before 
yesterday. Even then, what brought it to light was the inability to 
pass-through G.729 without licenses on the box (thanks to JerJer and 
drumkilla for bringing it to my attention). Nobody noticed that _all_ 
pass-through (but not natively bridged) calls were being turned into 
SLINEAR and back, rather than being left alone.