Can I offer a few notes about the Mastertouch record/editor system as
it uses a player piano to MIDI conversion as a vital component.  Please
excuse any of my loose terms or descriptions as I came into this field
from the computer and animatronics world, not the player piano world.

Switches under a MIDI keyboard need to sense the travel speed of the
struck key.  The intensity of the sounded note is calculated from this
velocity.  Mastertouch had a player piano already outfitted with simple
on/off switches, from an adaption done in the early 1960s which directly
controlled a sophisticated bank of relays operating the master punch machine.
The electronics from a stripped down Roland 88-note MIDI piano was spliced
into this keypad, so the output could be captured into a MIDI sequencer
programme on a PC.

The next stage is the custom MIDI to 88-hole conversion software.
After inputting the required holes per beat and beats per bar parameters,
the software calculates from the tempo of the MIDI track to work out the
roll speed and the "quantizing".  In MIDI terms, the holes are like
quantizing to a very coarse resolution.  The MIDI velocity is ignored
as 88 note rolls are either on or off.  To adapt it for artist rolls
would need a complete replacement of the keyboard switches, and software
to interpret the individual key velocities into Recordo or Duo-Art or
whatever control formats.

A couple of problems crop up when copying from old rolls compared to
playing new arrangements.  With live playing the musical beats are aligned
with the MIDI bar marks by playing in time with the MIDI metronome function.
When playing from a roll there is no way to automatically align the MIDI
bar start code with the correct moments of the music.  The slight stretch
and shrink of the paper due to humidity etc., and the variations in the
speed of the take up spool would make it a task for "artificial creativity"
beyond the scope of artificial intelligence.  Our pianists have worked out
how to handle these recordings quite well.

When we experimented with recording from the hammers or the sound we
found the delays between the key strike and follow on actions and the
resultant sound added shifts in timing.  MIDI and electronic response
times are in the thousandths of a second, and are locked predictably.
The response times for keyboard to hammer to live sound is very different.

We also experimented with a light sensitive track bar with software to
realign the tiny shifts in time between holes which should start or
finish in sync with each other.  It's a motorised thing with a shaft
encoder sensing the actual travel of the paper.  Each row of holes gets
scanned at a resolution of less than 0.01" while the software does
an xor, an and & an or back and forth across a hundred bit word to do
the realignment.  The correct placement of burred and slightly tattered
holes can be reclaimed with this.  We could even regenerate a master for
2:1 and 3:1 strings of holes.  The development had to be put aside at this
stage because of lack of funds, but it works.  We roughed out a few
algorithms so that the source roll could be almost any format, and the
final roll could be a different one.  Interpretation of the holes and
reformatting can go in software from the 100-bit word.

If there is enough interest we could look into restarting the project,
but as someone else pointed out, it's definitely not a trivial task.
With our normal way of budgeting it would be less than aus $10k, as the
basis of the system has already been researched as a feasibility project.

Sorry that there's some parts I can't (won't) describe, as it's part
of the system developed exclusively for Mastertouch.  Incidentally, it
was written in Forth and assembly to run on an old XT, and only needed
slight adaption of the timing for MIDI playback of the rolls in order
to run on an AT, a 486, and more recently on DOS within Windows 95.
I'd hesitate about rewriting it for windows as control of the exact
timing of the computer's hardware and interrupts is more awkward.
The experimental light sensitive track bar rig had a 4 MHz 6502 Forth
controller passing data to a PC for the more relaxed logic functions.

The older relay system can still be switched in.  As I worked out the
computerised system I realised the modern computer was taking part in
just one more event in a long history of leading edge technology, from
the era of vacuum bellows control through late 1950s relays.  It's a
pity there's no valve electronics in it.  It also links in with a tape
recording Pianocorder system with an added MIDI in/out adapter box.

Gary Luke   feraltek@zeta.org.au   fax +61-2-9519.9907