[Clam-devel] CLAM_OPTIMIZE bug fixed

[Hernán Ordiales]

On 9/7/07, Pau Arumi <parumi at iua.upf.edu> wrote:
> En/na Hernán Ordiales ha escrit:
> > On 9/6/07, Pau Arumi <parumi at iua.upf.edu> wrote:
> >> En/na Hernán Ordiales ha escrit:
> >>> On 9/5/07, Hernán Ordiales <h at ordia.com.ar> wrote:
> >>> [snip]
> >>>>>> After some testing I reverted a change on FundFreqDetectConfig that
> >>>>>> created artifacts when setting
> >>>>>> the CLAM_OPTIMIZE flag. All other optimizations sound good to me but I
> >>>>>> am testing offline (non-Network mode)
> >>>>>> so further testing may be needed.
> >>>> nice, i will check how sounds now
> >>> yes, definitely sounds much better now (tested at NE with harmonizer
> >>> and pitch-shift networks)
> >>> (optimize_and_lose_precision flag)
> >> example online?
> >
> > ok :-)
> >
> > here you have both versions pitch-shifted with values 1 , 1.3 and 1.5:
> > http://h.ordia.com.ar/tmp/clam/index.php?folder=Y29tcGFyaXNvbg==
> >
>
> But still sounds with artefacts and bubbling effect in contrast to
> the SMSTools one. Both versions (default, optimize) seems very
> similar to me.

yes, were different things/bugs

that's the story: at first, we suspect that the bug came for cpu load
(because were first noticed at harmonizer which uses many
pitch-shifts), then i was trying to reduce that, and in other things i
tried with this optimize flag, but i also noticed (in addition to the
bug) a very poor sound quality compared with default version... this
was very weird to xavier and now he solved this issue. Now default
version is _comparable_ in sound quality with optimized one (with
trigonometric tricks and less cpu load) again (and first time for me)

> I think we should need to compare the processing done in SMSTools
> --segment wise-- and the one in streaming --independent frames.

yes, agree

as soon i discovered that the bug came for pitch-shift i tried with
SMSTools one to check and was working (was an audible test), now if i
can't find any NE version working i will look at there... (source code
now)

cheers,
-- 
Hernán
http://h.ordia.com.ar
GnuPG: 0xEE8A3FE9