[maemo-developers] Signal processing on N800

From: Stefan Kost ensonic at hora-obscura.de
Date: Sat May 17 19:43:39 EEST 2008
Hi,
Sarah Newman schrieb:
> Do you think that is significantly more efficient than kiss fft?  We 
> have been considering using FFTs optimized for arm but platform 
> independence is always nice.
 >
GStreamer has a copy of kiss-fft included. Would be nice to know how it compares 
  to the CMU one.

Stefan

> 
> David Huggins-Daines wrote:
>> On Wed, 2008-04-23 at 14:15 -0500, nisha jain wrote:
>>
>>> to port it to N800 later on...Please let me know if some one has
>>> developed such 
>>> application and used fft calculation for discrete signals... 
>> If you are dealing with real-valued data (e.g. audio), you can use the
>> fixed-point FFT routine in CMU Sphinx if you like.  It's not
>> specifically optimized for the ARM but it's pretty fast.  Browse the
>> code at:
>>
>> http://cmusphinx.svn.sourceforge.net/viewvc/cmusphinx/trunk/sphinxbase/src/libsphinxbase/fe/fe_sigproc.c?view=markup
>>
>> The functions you need are fe_create_twiddle() and fe_fft_real().  They
>> come in 16 and 32-bit versions, based on the FIXED16 preprocessor
>> symbol.  The 32-bit version is a bit confusing since it uses 20.12
>> fixed-point format, so I'd recommend the 16-bit one (which is faster
>> anyway).
>>
>> You'll have to do a bit of work to separate it from the rest of the
>> front end processing, but that shouldn't be hard.
>>
>> Better would be to use TI's dsplib on the DSP core, but I still haven't
>> gotten around to figuring that out.
>>
>> _______________________________________________
>> maemo-developers mailing list
>> maemo-developers at maemo.org
>> https://lists.maemo.org/mailman/listinfo/maemo-developers
>>
> _______________________________________________
> maemo-developers mailing list
> maemo-developers at maemo.org
> https://lists.maemo.org/mailman/listinfo/maemo-developers


More information about the maemo-developers mailing list