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 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?

The Ampico entry at http://pianola.org mentions that technicians coding
the expression information would have been using tables and algorithms
to determine the appropriate coding.  This of course would have been
tweaked before finalizing the master.  But I see no reason why one
could not design an algorithm to translate peak MIDI velocities into
the necessary expression coding with some measure of authenticity.  The
program would not only need to determine the coding based on single
notes, but also the effect on the vacuum supply of several simultaneous
notes, and their effect across both the left and right divisions as
well, as is described in the comments on the beginning of the roll
found here:

    http://members.socket.net/~rtaylor/coding_master_roll_72083.html  

This is certainly a feat well within available technology, provided one
finds a capable programmer.  It would open the possibility of faster
production of new paper expression rolls and of digital e-rolls.  Also,
when used in conjunction with a piano equipped with the e-valve system,
the algorithms could allow a performance recorded in MIDI format to be
played directly and immediately from the file on an original pneumatic
instrument, without the need to convert the file to an e-roll.  This
would also mean one could easily translate and produce the same roll of
the same performance in any expression format that the program could
handle, Ampico, Duo-Art, Welte, Red Welte, what have you, provided you
have the proper algorithms for each system.

I welcome your thoughts on this subject, as it is one that has been
churning around in my mind for quite some time now.

Ben Willis