Re: [iiwusynth-devel] iiwusynth

[Tim Goetze]

M. Nentwig wrote:

>Hello,
>
>I'll leave most answers to Peter, but I can at least try to help with
>the two last items:

thanks!

>- The velocity mapping has been changed roughly three days ago. I
>remember, that soft notes dropped too low, because of the shape of the
>concave-convex curves (this is now implemented according to the SF2.01
>specs, everything else is a bug).

good to know that the spec is being followed in the engine. it's
easier to direct counter-measures that way (mapping velocity or
modifying the soundfonts).

>If a known sound font doesn't work as it should, please let me know. The
>specs are rather vague, especially when it comes to the vel-to-filter
>default modulator.

for me it's hard to tell how they *should* sound -- first, my awe64
came with only 512k so most fonts don't work with it (plus i assume
the emu8k chip uses cheap integer calculation internally anyway). and
second, i don't have any other sf2 engine at my disposal.

[...]

>- The gain is frequently causing me a headache. But the default should
>be safe, that is low.
>So I have changed the default from 0.9 to 0.2 (dropped roughly by 13 dB)
>in the last changes to CVS. This will make it even worse for you.
>Luckily, you can simply issue the command
>gain 1
>from the command line,  iiwu_synth_set_gain(synth, gain) should do the
>same from C.
>Sadly, most soundfonts on the net are not normalized, so some gain boost
>is needed (of course it will also lift up the quantization noise at the
>low end).

the 2 MB font that came with the awe is normalized, but the output
peak varies from preset to preset. it never seems to exceed 0.2f
though, so at least it can be said to work as expected. :)

>For real-time piano sounds I usually use a limiter as LADSPA plugin to
>prevent the worst digital clipping, and then as much gain as sounds
>good.

yeah, the limiter is a good idea to help out here. however in my
project the synth output signal will feed a DSP chain, and it's far
more flexible to apply the limiter somewhere in that chain than to
have it at a fixed place within the synth.

therefore the code clamping the output to [-1, 1] in *_write_float is
a major annoyance to me -- it effectively makes the signal unusable
once it starts to oversteer within the engine. i have a feeling this
clamping is based on the assumption that the signal already went
through a limiter.

btw, is somebody working on making drumkits work right? in the
above-mentioned 2 MB font, a closed hihat should cut off an open
hihat. with the awe64 it does, in iiwu it doesn't.

tim