Speex dev - Mar 2007 - Performance of the acoustic echo canceller

Hi All, I am working on my master project to port the Speex Acousic Canceller to
FPGA. But I am not able to get a satisfied result in its performance whenI
integrated it to a SIP softphone as my first step.  The SIP softphone calls to a
PSTN gateway and lands to my land line phone.  I can hear the echo of my every
word while I talks on the land line phone.   I am using G.711 codec, frame_size
80 and both 1024 and 2048 tail length. The code was compiled as floating point
and fixed point in the PC. But noneof them help. I made sure the the
synchronization between echo_frame and input_frame was stable and the delay is
consistant about 10ms. The echo return lossis around 20db but output_frame and
input_frame signal is about the same.  I even tried Jean-Marc's latest mdf.c
from the subversion but no luck. Has anyone test the performance in any other
softphones and how well does it work?   I have the recording files, anyone would
be interesting in taking a look? I wonder if there is something wrong with my
input, setup...BTW: did anyone ever try to port the algorithm to hardware?  I
will do some profiling and implement the most compuation expensive function in
the Xilinx FPGA.I would really appecriate if anyone can share some  experience
or provide adavise and suggest!  I like to contribute my work to the Speex
project if my project turns out to be good end.  Regards,William
_________________________________________________________________
Connect to the next generation of MSN Messenger?
http://imagine-msn.com/messenger/launch80/default.aspx?locale=en-us&source=wlmailtagline
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.xiph.org/pipermail/speex-dev/attachments/20070305/271bf36a/attachment.html

Best advice I can give is to record the near-end and far-end files, pass
them through testecho (disable the preprocessor) and check the result.
Then check why it's not working (if that's the case). Some usual
suspects:

1) Clock drift between capture and playback
2) Bad synchronisation
3) A non-linearity somewhere (bad speakers, noise gate, ...)

If you try everything and can't find what the problem is, please upload
the files somewhere and send an email with the links (do not email the
files directly).

	Jean-Marc

William Zhang a ?crit :
> Hi All, I am working on my master project to port the Speex Acousic
> Canceller to FPGA. But I am not able to get a satisfied result in its
> performance whenI integrated it to a SIP softphone as my first step.
> The SIP softphone calls to a PSTN gateway and lands to my land line
> phone.  I can hear the echo of my every word while I talks on the
> land line phone.   I am using G.711 codec, frame_size 80 and both
> 1024 and 2048 tail length. The code was compiled as floating point
> and fixed point in the PC. But noneof them help. I made sure the the
> synchronization between echo_frame and input_frame was stable and the
> delay is consistant about 10ms. The echo return lossis around 20db
> but output_frame and input_frame signal is about the same.  I even
> tried Jean-Marc's latest mdf.c from the subversion but no luck. Has
> anyone test the performance in any other softphones and how well does
> it work?   I have the recording files, anyone would be interesting in
> taking a look? I wonder if there is something wrong with my input,
> setup...BTW: did anyone ever try to port the algorithm to hardware?
> I will do some profiling and implement the most compuation expensive
> function in the Xilinx FPGA.I would really appecriate if anyone can
> share some  experience or provide adavise and suggest!  I like to
> contribute my work to the Speex project if my project turns out to be
> good end.  Regards,William 
> _________________________________________________________________ 
> Connect to the next generation of MSN Messenger 
>
http://imagine-msn.com/messenger/launch80/default.aspx?locale=en-us&source=wlmailtagline
> 
> 
> 
> ------------------------------------------------------------------------
> 
> 
> _______________________________________________ Speex-dev mailing
> list Speex-dev@xiph.org 
> http://lists.xiph.org/mailman/listinfo/speex-dev