I quite agree that a key product of roll scanning is new music rolls.
Luckily, this is also a major aim of the Rollscanners project.  Because
it's a joint project there are many aims from it - but as it's been
thought through carefully it can meet all the various objectives.

No matter what you need to do with a roll, the first thing you have to
do is scan it into the computer.  After that all the manipulation is
in software, so in principal if you can think of it you can do it!  The
secret of good design is to make sure that at no point do you introduce
any unnecessary restrictions.

Scanning gives you a fairly large graphical image.  From this you have
to work out where the perforations lie across the roll, dealing with
tracking issues, and then work your way along the roll to find the
start and end of each perforation.  As this is related to reading the
physical roll, it is common to whatever you want to do with the
information later.

After this, if you want to pursue the 'holy grail' of recreating the
original roll master, some sophisticated processing is required to
identify each individual perforation in the original roll.  This stage
has proved hard so far for practical use, as excessive manual
intervention has been required, but better software seems to be
progressing well.

Because the master recreation process removes scan errors and paper
skews, etc, the result can be regarded as an infinite-resolution scan
of an undamaged roll, with some additional punch-rate data.  Anything
that can be done with the original scan can be done better with the
recreated master!  (This is a very important point that I don't think
is widely understood.)

On the other hand, if you merely want an analogue scan of the
performance on the roll, you can simply strip out the bridges and any
specks without identifying the punch positions.  This is far simpler
and is currently the norm.

If you have a suitable roll perforator, you can make an exact replica
of the original roll using the recreated master, or significantly less
accurate copies from the de-bridged analogue scan (because your new
perforator steps won't match the original and will introduce readily
discernable timing errors even if scanner and perforator are high
resolution ones).  Hopefully the analogue copied rolls will be made a
thing of the past fairly soon, as it'll be a shame to waste time and
effort in producing low-fidelity products when perfectly accurate ones
cost exactly the same to make!

Either the recreated master roll (de-bridged) or the analogue
interpretation can be used directly for operating in-line valves on
a player.  If you care to, you can interpret the slots as MIDI 'note
on' and 'note off', and use any expression coding to set the MIDI note
dynamics and hence simulate the original performance.

These MIDI files really should be created on demand from the roll
master, so that the MIDI parameters can be set up for the playback
device being used.  I cannot bring myself to regard the MIDI files
as anything other than consumable items -- they are certainly not
a suitable archival medium for roll performances! [*]

All these products have their place: recut rolls are obviously the
prime focus for most instrument owners, in-line valves have an obvious
appeal for those who want to pay for the convenience and like the music
rather than the paper medium itself, and the MIDI files for
computer-playback are ideal for identifying pieces you'd like to hear
played properly on a real piano.  The Rollscanners group gives us all
three, and hasn't been unnecessarily restricted to any one interest
range.

To further impress, the technology and (most of) the software is
relatively straightforward stuff (see pictures of some minimalist roll
scanners on Terry Smythe's web site!).  The capability to do all this
lies within many people's ability and pockets -- for which we have to
thank pioneers like Richard Stibbons who have ploughed their way
through ever-simpler devices to get where we are now.

Julian Dyer

 [ * The data quality isn't related to the data storage format.
 [ The "Standard MIDI File" (SMF) data format is routinely used to
 [ store precise data for replicating music rolls, as well as for
 [ storing imprecise transcriptions.  Don't blame MIDI -- it's simply
 [ "garbage in garbage out!"  -- Robbie