Re: [Iaxclient-devel] speex echo cancellation tested?

[Matthew C. <mc...@no...>]

On Wed, 2008-04-23 at 23:46 -0400, Mihai Balea wrote:
> On Apr 23, 2008, at 6:06 PM, Peter Grayson wrote:
> 
> > Hi Matt,
> >
> > On Wed, Apr 23, 2008 at 4:39 PM, Matthew Crane <mc...@no...>  
> > wrote:
> >> On Wed, 2008-04-23 at 13:57 -0600, Matthew Crane wrote:
> >>> Hello all:
> >>>
> >>> I'm investigating how well the speex echo canceller works in the 2.1
> >>> branch of iaxclient.  Whenever I turn on the speex AEC in our  
> >>> client,
> >>> the audio that is sent by the client.
> >>
> >> Sorry... that last sentence should read, "...the audio that is sent  
> >> by
> >> the client is almost completely unintelligible."
> >>
> >>> Has anyone tested the performance of the speex echo canceller with 8
> >>> kHz-sampled audio?
> >
> > I've got a patch that's been sat upon for quite a while now that
> > begins integrating the new (or at least greatly improved) speex aec
> > module (introduced in speex-1.2beta2) into iaxclient. I've been
> > holding off committing these changes because the results have really
> > only been verified to be good on Mac. However, I think this code goes
> > in the right direction. Even when the AEC fails to cancel echo, it
> > doesn't make things sound worse.
> 
> You can actually find this code in branches/team/mihai/echocan.
> As Pete said, performance is pretty good on Mac.  On windows it seems  
> that it does not work, but at least it doesn't make things worse.
> The reason seems to be related to crappines in Windows' audio APIs.   
> Performance on Linux has not been tested (at least not by me)

> Anyways, bottom line is the code currently in trunk does not have  
> functioning AEC
> 

Thanks Pete and Mihai.  I'll have a look at the branch.  For some
reason, I assumed that the requirement for speex-1.2beta3 in the 2.1beta
of iaxclient was because of AEC.  :-)

If I can verify that the latest speex AEC works well inside of iaxclient
on linux, I might try getting one of the low-latency Windows portaudio
implementations working.  In regards to the wmme API, you're right --
it's unusable for AEC.

-Matt