Robbie wrote:

> Howzat again?  Yes: true random noise on the time base doesn't sound right,
> either.  Therefore, the original performer was NOT random in his timing,
> and there must be some sort of correlation with the musical figure he's
> performing.
>
> Artis would welcome constructive advice on this task.  Any ideas out there?

I met couple of people studying piano performance rules and applying
them to the synthesis of piano performances. There is a group at KTH
Stockholm (Royal Technical University) lead by Prof. Johan Sundberg
who did a lot of work formulating performance rules. They implemented
a rule-based system for playing contemporary piano pieces by a
computer. Their work has been published (among other places) in the Computer
Music Journal (I'm not sure about the number, but should be around
1991-92, I can check it at home). There is also a guy called Roberto
Bresin from Italy who implemented their rule system using a neural
net approach. His goal was to teach the network to adapt to a
specific player's style. This is something that may be needed in our
case. I met Roberto a couple of times at various conferences, and I
heard his demos. Well, the problem seems to be quite difficult, so
don't expect a ready-made, fully working solution.

Btw., several sequencers have a 'humanize' feature. I don't know how
good or bad these are, but certainly they try to imitate the human
player somehow.

I will collect the references I have and post them later to the list.

Back to the WAV to MIDI conversion, I'm very much interested in this
thread as well. I found the GA idea quite interesting. Let's keep
talk about it!

Zoli¶
-----¶
Zoltan Janosy <janosy@hit.bme.hu>¶
-----------------------------------------------¶
= (Mr.) Zoltan Janosy <janosy@hit.bme.hu>¶
= Technical University of Budapest¶
= Department of Telecommunications¶
= Sztoczek u. 2. 308., H-1111 Budapest, Hungary¶
= Tel: +36-1-463-2093, Fax: +36-1-463-3266