Speex dev - Dec 2006 - problem with echo cancellation

Hello Jean-Marc,

 

I solved the variable delay problem, but I still have trouble with
speex_echo_cancel(). When i try testecho.c with clean speech for far-end input
and same speech with attenuation, a bit of reverb and 50-150 ms delay, all this
done in sound editor, for mic input, i get 5-8 db attenuation.

 

But when i use the same speech played and recorded for mic input, i see about 5
db of attenuation only if i remove the delay (150ms) at the beginning of mic
input file to make far-end and mic input files to go almost ?in sync?, and even
in this case there?s almost no attenuation during about a second after pauses in
speech. Looks like the filter re-adapts after each pause. When the delay is at
least 50 ms, the output of echo cancellation doesn?t differ much from mic input.

 

I use speex1-2beta1 version, and i applied the patch you suggested. I use a
sampling rate of 8000 which i have set with speex_echo_ctl(),  a frame size of
128 and tail length of 2048. I have tried other tail lengths (640, 1024, 4096)
with no significant effect. I use the same soundcard for capture and playback,
and quality speakers and microphone to avoid nonlinear distortions.

 

How much attenuation can the echo canceller normally produce? Is there anything
that i'm doing wrong? Getting this to work is very important for me,

 

Thanks a lot!

 

Julia

Hi Julia,

Version 1.2-beta1 has a bug in the echo canceller. Try either svn or
1.1.12 -- or apply this patch to 1.2beta1:
https://trac.xiph.org/changeset/11882

	Jean-Marc

julia rg a ?crit :
> Hello Jean-Marc,
> 
>  
> 
> I solved the variable delay problem, but I still have trouble with
> speex_echo_cancel(). When i try testecho.c with clean speech for far-end
input
> and same speech with attenuation, a bit of reverb and 50-150 ms delay, all
this
> done in sound editor, for mic input, i get 5-8 db attenuation.
> 
>  
> 
> But when i use the same speech played and recorded for mic input, i see
about 5
> db of attenuation only if i remove the delay (150ms) at the beginning of
mic
> input file to make far-end and mic input files to go almost ?in sync?, and
even
> in this case there?s almost no attenuation during about a second after
pauses in
> speech. Looks like the filter re-adapts after each pause. When the delay is
at
> least 50 ms, the output of echo cancellation doesn?t differ much from mic
input.
> 
>  
> 
> I use speex1-2beta1 version, and i applied the patch you suggested. I use a
> sampling rate of 8000 which i have set with speex_echo_ctl(),  a frame size
of
> 128 and tail length of 2048. I have tried other tail lengths (640, 1024,
4096)
> with no significant effect. I use the same soundcard for capture and
playback,
> and quality speakers and microphone to avoid nonlinear distortions.
> 
>  
> 
> How much attenuation can the echo canceller normally produce? Is there
anything
> that i'm doing wrong? Getting this to work is very important for me,
> 
>  
> 
> Thanks a lot!
> 
>  
> 
> Julia
> 
> _______________________________________________
> Speex-dev mailing list
> Speex-dev@xiph.org
> http://lists.xiph.org/mailman/listinfo/speex-dev
> 
>

Jean-Marc Valin ?rta:
>> How much attenuation can the echo canceller normally produce?
Take a look at Example 6 - "Speex in a Disco" here:

http://www.surfnonstop.com/~bandit/speex/echocard1/

Andras