Re: [iiwusynth-devel] Filter: Implemented, fixed

[Josh Green]

On Tue, 2002-04-02 at 03:17, Christian Nentwig wrote:
> Hello,
> 
> found it. I hadn't tested sounds, which change the filter frequency over
> time: recalculating the filter coefficients produced some
> 'zipper'-noise. But please let me know, if it still misbehaves.
> 
> The updated version is again at
> http://prdownloads.sourceforge.net/millisampler/Filterimplementation_tar.bz2
> .

Tried it. Sounds much better now. Seems to be a little more CPU
intensive than the older one (haven't tested this thoroughly though),
but does sound more true to sound font standard. It would be nice to
compare an SB Live! or something in windows to the sound of iiwusynth (I
don't quite trust the Linux wavetable drivers). I did notice that very
fast increases in filter cutoff (say in the modulation envelope release
stage, with a sufficiently long volume release) causes a bit of filter
instability (loud scratching sounds or noisy stuff) even when sample
volume has completely died out, maybe thats whats supposed to happen?
Probably not.

> There is also a sound font test file, which sweeps the filter to the
> limit.
> Patch 00: Slow filter sweep
> Patch 01: Fast filter sweep
> Patch 02: Very fast filter sweep ('typewriter-sound')
> 

Cool, it was made with Smurf :)

> Speaking of it, there is something odd about the fcenter-information
> that is produced by iiwusynth: A linear sweep (cents per second) should
> reduce the frequency on a logarithmic scale. But the frequency change
> seems to be linear (it reaches 0, that's impossible when sweeping on a
> logarithmic scale).
> 
> Currently the filter coefficient calculation happens every IIWU_BUFSIZE
> (64) samples once for each voice, and it includes a sin(x) and cos(x).
> Has anybody an opinion on how CPU-intensive those are? Does it make
> sense in the days of FPUs everywhere to use lookup tables?
> 

I haven't looked at the clock cycles consumed by FPU instructions for a
while, but I do remember sin/cos being fairly costly. I'm not very
familiar with iiwusynth's synthesis code, but it seems like it would be
cool to make it somewhat pluggable. So you could choose different
algorithms for the various effects for speed/quality trade offs. It
would be real neat to have a non-realtime file rendering mode for
iiwusynth. Then, once you get your MIDI files to your liking, use all
the best effect modules and just render to a file :)
Anyways, dreaming away. All this sounds good of course, but implementing
it is another thing. I'm currently too busy with Swami to lend much
support to iiwusynth, so..

> Regards
> 
> Markus
> 
> PS: Nice sound font. My current favourite: El-Cheapo Organ :-)
> 

Yeah its a nice sound font. I like how its all based on really simple
waveforms, showing that soundfonts can be used for analog type stuff.
Cheers!
        Josh Green