In the 080810 MMDigest, Ben Willis wrote:

> I happen to have a jazz pianist friend who shares my enthusiasm for
> player pianos and other mechanical instruments.  On several occasions
> he has told me that one of his goals as a performer is to cut his own
> piano roll or series of rolls, preferably in one of the more common
> expression formats.

I would encourage your friend to also consider recording some albums
for the modern solenoid-based player systems (Yamaha Disklavier,
PianoDisc, QRS Pianomation, Pianoforce, Live Performance).  There is
a shortage of well-played, non-cheesy material for these systems,
giving them a bad reputation, and there is a larger audience for new
recordings.  Tools are available to author content for these systems
in both diskette and CD formats ( http://www.giebler.com/ ,
http://dp70.dyndns.org/mid2pianocd/ ).

> I know that the expression coding for new rolls has historically been
> worked out by hand, a long and laborious process.  But reading about
> and listening to efforts to emulate expression rolls in MIDI format
> got me wondering: Has any work been done to reverse the process?  Has
> anyone tried creating new expression rolls directly from MIDI-recorded
> performances?

I have dealt with some of these issues in transcoding MIDI files to
play on the Marantz Pianocorder, a solenoid-based player system
designed in the 1970s which bears some resemblance to the pneumatic
reproducing systems.  The Pianocorder implements split stack expression
with a configurable split point that can be set to match the Ampico,
Duo-Art and Welte standards (or even changed as needed during the
course of a song).  Each half of the keyboard may be set to any of 32
expression levels.

Encoding material to Pianocorder format involves compensating for the
variable delay in solenoid strike times at various expression levels.
It also involves some heuristics to determine which notes struck around
the same time should most strongly influence the expression levels for
each half of the keyboard.  The encoder also makes adjustments to the
note timing to account for mechanical limitations of the piano and to
reduce aliasing resulting from the limited time resolution of the media
(1/35th of a second for Pianocorder, analogous to the punch step of a
paper roll).

In contrast with pneumatic reproducing systems, the Pianocorder can
adjust its treble and bass expression to any level within 1/35th of
a second, and there are no vacuum reservoir issues to worry about.
Some adjustments are needed to counteract the effects of solenoid
heating during longer performances.

I believe that some of the techniques used in encoding to Pianocorder
format could be adapted to encoding for the pneumatic systems.  There
are surely a number of MMD readers qualified to develop such an
encoder, but the process of refining the encoding software for each of
the target systems would be substantially easier with extended access
to a well-maintained pneumatic piano equipped with e-valve control.

It has occurred to me to extend MID2PianoCD to perform pneumatic
encoding in addition to the modern formats it already supports; I think
Spencer Chase and I even discussed this a while back.  But I have no
local access to e-valve-equipped pianos and, with young children at
home, I'm presently short on time as well.  But I'll continue
collecting information towards eventually tackling this problem.

Mark Fontana