Re: [Linuxsampler-devel] [PATCH] fine tuning of the velocity curves

[Andreas P. <and...@lk...>]

Christian Schoenebeck skrev:
> Es geschah am Samstag 20 November 2004 18:01 als Andreas Persson schrieb:
> 
>>Hello,
> 
> 
> Hi Andreas!
> 
> 
>>This is my first post to this list, and my first attempt to contribute
>>to this impressive project. What I have done is some rather
>>overambitious measuring and fine tuning of the velocity curves.
> 
> 
> So you actually measured Gigasampler's original velocity curves?
> 

Yes, I did a lot of measurements of GS 2.5. Input was a minimal gig with
a sawtooth sample and a MIDI sequence of the same note repeated with all
127 different velocities. Output was a wav, which I run through a small
program to get data for a velocity x volume graph. I did this for all 15
curves with scaling set to 20, and for some curves I also varied the
scaling parameter.

>>The three functions for linear, nonlinear and special are replaced with
>>15 separate curves, approximated with line segments. The scaling
>>parameter is also handled a bit different than before.
> 
> 
> Do you think it might be worth to approximate a polynomial function instead of 
> using linear segments? Btw here

When looking at the graphs for the measured nonlinear curves, it seems
as GS also uses linear segments. For the special curves  I'm not so
sure, perhaps some of them or some segments of them could be replaced
with polynomials or perhaps exponential functions, but I doubt it would
be worth it.

I did try to find functions, I found this site for function fitting to
be great: http://zunzun.com, but I did not found any functions that
could approximate whole curves well in a consistent way.

>  http://www.linuxsampler.org/doc/engines/gig/velocitycurves.pdf
> 
> is a graphical comparison betwenn the original measurement curves made by Mark 
> Knecht and the calculated functions currently in use in LS/libgig - just in 
> case somebody's interested in a graphical comparison of the curves.

Uh-oh, It seems as somebody else than me has spent a lot of time and
effort on this, I didn't mean to step on any toes... I started doing
this tuning when I noticed that my favorite piano gig was not feeling
right, It was not audible at all when playing soft. It uses curve type
nonlinear, range 1 and scaling 0.

Here's a plot of a measured nonlinear-1-20 curve together with my
line-segment-approx and the original linuxsampler function:

http://hem.spray.se/andreas56/plotfile.png

> Has somebody already tested Andreas' patch? Do the new curves feel more 
> accurate? Mark perhaps?

Well, I have :). Apart from some artificial tests, the piano feels
better with the patch. Other than that I have not tested the feel that much.

> CU
> Christian

/Andreas