19 Oct
2011
19 Oct
'11
4:37 p.m.
On Wed, Oct 19, 2011 at 03:23:34PM +0200, Jeremy Jongepier wrote:
Soundwise, as far as I have understood as a DSP layman
it could
introduce artefacts, but I wonder in how far those are audible. Guess if
you're a DSP layman you probably can't distinguish any possible
artefacts. The other point is the DSP load of JAMin, last time I used
JAMin on my quad core machine I recall the DSP load rose about 20 to
25%. If I use separate plug-ins it's about 10%. But I have to test that
again to be sure sure.
The two issues are related. The FFT based EQ in Jamin uses a form
of block processing that leads to a filter that is not time-invariant,
it produces AM on some frequencies. In the first release that was very
obvious, you could actually hear it quite easily. Instead of changing
the algorthm to a correct one (which would be quite similar) the Jamin
devs chose to mitigate the effect by increasing the overlap between
successive blocks. IIRC there are now 32 overlapping blocks at any
time. This reduces the modulation to acceptable levels, but is also
what is responsible for the high CPU load. A correct implementation of
this type of EQ requires an overlap of half the FFT size, so it would
require much less CPU.
Ciao,
--
FA