[Asterisk-Users] Fun with Echo

[James Harper]

> There is a spec for echo cancellation on PSTN called g.168.  I believe
> it's a 
> suite of tests which put the echo canceller through its paces and if
you
> pass
> them you are certified to conform to g.168. None of the echo
cancellers in
> zaptel conform to this, whereas the Octasic, Tellabs and other
hardware
> echo
> cancellers all do.  If someone were to put the effort and energy into
> making
> the software echo cancellers compliant, you should find similar
results to
> the hardware echo cans.

Just out of curiosity, what form does the testing take? Is it simply a
set of audio sample pairs (outbound, and inbound with some echo of
outbound) that you run the ec against and the measure the results?

What metrics are measured? I can only think of:
. the amount of echo left after the ec has run
. training time

How hard would it be to make a g.168 test bed that the Asterisk ec's
could be run against?

I can definitely see the value in even just making some audio samples
for people to listen to and giving each ec (with version number and
tunable parameter values) a score against each sample, even if it's just
a subjective score based on perception from users. That way a newbie
could listen to the audio samples to find one that sounds like the sort
of echo they are getting (mainly based on delay I guess but different
types of echo will have different frequency responses than others) and
then can pick a good ec based on score.

Comments?

James