Speex dev - Aug 2005 - Echo cancellation questions

Hi!

I have a few questions considering echo cancellation algorithm in Speex:

1. In the manual it's stated that the delay between the input signal and the
echo signal must be small. How small should it be? Is for example 100 ms
acceptable?

2. Does echo cancellation algorithm deal well with situation, when one of the
users of Speex-based VOIP application has "record-what-you-here" sound
card mixer option selected? Did anybody test the algorithm in such conditions?
In an application I'm working on the result is not nice - both users
(especcialy the one with mentioned option enabled) hear periodic (approx. once a
second) loud  noise ("knocks"), which seems to be an effect of
amplification of background noise caused by echo. Am I using speex_echo_cancel
improperly or it's not possible to eliminate echo in such conditions?

3. Does e.c. algorithm works as good in wideband as in narrowband mode?

I'm using Speex 1.1.10 library.

-- 
Greets,
Mikolaj

----------------------------------------------------
Jakie jest Twoje IP i jaka jest rzeczywista szybko?? Twojego ??cza?
http://klik.wp.pl/?adr=www.TwojeIP.wp.pl&sid=469

> 1. In the manual it's stated that the delay between the input signal
> and the echo signal must be small. How small should it be? Is for
> example 100 ms acceptable?
100ms would probably work, but less is always better. If you know it's
100ms, all you have to do is delay the far end signal.
> 2. Does echo cancellation algorithm deal well with situation, when one
> of the users of Speex-based VOIP application has
> "record-what-you-here" sound card mixer option selected? Did
anybody
> test the algorithm in such conditions? In an application I'm working
> on the result is not nice - both users (especcialy the one with
> mentioned option enabled) hear periodic (approx. once a second) loud
> noise ("knocks"), which seems to be an effect of amplification of
> background noise caused by echo. Am I using speex_echo_cancel
> improperly or it's not possible to eliminate echo in such conditions?
The problem is that the feedback loop (with gain > 1) builds up before
the echo canceller has time to adapt. 
> 3. Does e.c. algorithm works as good in wideband as in narrowband
> mode?
Should work, haven't tested it.

	Jean-Marc