[Speex-dev] More integer / ARM-specific optimizations

ndno72-speex at yahoo.com ndno72-speex at yahoo.com
Tue Jun 28 15:36:24 PDT 2005


Hi Jean-Marc,

Thanks for the quick response.  Here are the numbers
for our little ARM systems:

CPU: ARM920T, 200Mhz 
Speex specs: Complexity 2, 5.95Kbps
Encoding: ~30%, Decoding: ~4%, 
  with Echo Cancellation / Pre-processor: > 85%
Compilation: arm-linux-gcc -pipe -Wall -W -O2
             arm-linux-g++ -pipe -fno-exceptions
-fno-rtti -Wall -W -O2
Configuration: --host=arm-linux --enable-fixed-point
--enable-arm4-asm

Andy


--- Jean-Marc Valin <Jean-Marc.Valin at USherbrooke.ca>
wrote:

> Hi,
> 
> There are long-term plans to have everything
> (including echo canceller
> and preprocessor) to fixed-point. However, I cannot
> say when/if that
> will happen. While the VAD/DTX should be too hard
> (even a decent float
> emulator will work), echo canceller and preprocessor
> would require quite
> a bit of work.
> 
> Just to have an idea of how Speex does, would you
> mind sharing the
> results you have on ARM? More specifically:
> -ARM version, speed
> -Complexity, bit-rate used
> -Percentage of CPU required
> -Compile flags/configure options
> 
> Thanks,
> 
> 	Jean-Marc
> 
> Le lundi 27 juin 2005 à 17:44 -0700,
> ndno72-speex at yahoo.com a écrit :
> > Dear Speex Developers,
> >  
> > We are currently using the basics features of
> Speex
> > 1.1.10 on our little embedded ARM processor and it
> > works great!  However, we hit a brick-wall when
> trying
> > to use echo cancellation or the pre-processor; our
> CPU
> > (no FPU) jumped to above 90%.  Are there any plans
> in
> > the immediate future to optimize the VAD/DTX and
> > pre-processor code of Speex (convert to
> fixed-point,
> > ARM optimization, etc.)?  I can convince the
> company
> > whose is sponsoring my project to donate some
> money
> > into your cause.  Thank you and keep up the
> wonderful
> > Speex work.
> >  
> > Andy
> >  
> > 
> > _______________________________________________
> > Speex-dev mailing list
> > Speex-dev at xiph.org
> > http://lists.xiph.org/mailman/listinfo/speex-dev
> 
> 
> 



More information about the Speex-dev mailing list