On Mon, 2013-06-10 at 10:07 -0700, [email protected] wrote:
> > With these instructions, we can collect results from a variety of
> > machines, and perhaps narrow down the factors that affect errors in
> > the bit exactness of the codec and modem.
> 
> Small differences in the bit-exactness of floating point calculation are
> not errors. 

Sometimes these tests do expose significant errors (for example due to a
coding or porting bug), and sometimes the errors are not small.  They
need to be examined before dismissal.  

There is no easy way to evaluate and compare Codec output speech quality
so we can not simply dismiss  small differences in the output without a
good reason.

It's a very useful technique I have used for 20 years of Codec and DSP
development - especially for porting work (say fixed point simulation to
real time, or converting the modem from Octave to C).  It stops bugs.

For the modem port I had some test scripts that evaluated the error
ranges, like 1E-4 between the Octave and C code.  For the STM port work
I have some code and Octave that dumps internal codec states to allow
finer grained comparisons, but it's not automated.

In this case (building using cmake versus autotools) it was a simple
sanity test which I performed several weeks ago which I asked Richard to
reproduce.  His results seem positive, so I'll try it again some time.

Cheers,

David




------------------------------------------------------------------------------
This SF.net email is sponsored by Windows:

Build for Windows Store.

http://p.sf.net/sfu/windows-dev2dev
_______________________________________________
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2

Reply via email to