Lee Revell wrote:
On Sat, 2006-07-08 at 11:37 -0700, Florin Andrei wrote:
On Fri, 2006-07-07 at 19:57 +0200, Frank Barknecht wrote:
As Dave P. wrote, if a real *groove* is wanted, editing a lot of stuff
like timings and velocities by hand and ear is the usual way to work.
Obviously one takes this hard and time consuming road, because just
adding gaussian or other randomization is not achieving satisfying
results for the goal at hand.
A potentially good but certainly expensive method would be to record
many expert human players and analyze their playing. Some kind of
patterns must emerge.
Then try and reproduce that in software.
How does Hydrogen's humanize feature work? It certainly sounds good.
Maybe that could be broken out into a standalone app?
Lee
Lee, I've just checked the code.
It uses a random offsets with gaussian distribution (just like my
humanizer0.0.2). There's more: the algorithm used is the same I used for
my project (found here: http://www.taygeta.com/random/gaussian.html).
But hydrogen source uses rand() as a random generator, when I've heard
that random() is much better (more uniform distribution) and in fact I
also changed rand() to random() from v0.0.1 to 0.0.2.
c.
--
www.cesaremarilungo.com
