Andy Taylor asks for details of my efforts to build a roll scanner.
I'm quite happy to share the information -- I've been at it for four
years on and off and still haven't got there.  Each time you dig
yourself out of one hole you find yourself in an even deeper one !
The details would fill a book, but here's an overview.

The basic software steps are :-

  1) Scan the roll to produce a computer image.
  2) Clean the image to get rid of specks and oil spots on the paper
     (mostly done automatically in software)
  3) Search through the image horizontally to find the note tracks.

There are now two options. To create a MIDI file :-

  4) Strip out the webbing.
  5) Shift the timing of certain elements such as Duo-Art expression
     tracks.
  6) Build a MIDI file, interpreting expression and pedaling in the
     process.

or, to re-create the perforator master :-

  4) Search through the image vertically to find the perforator matrix
     (rather like a sophisticated form of quantizing).
  5) Build a perforator file.

The software looks complicated but is actually fairly straightforward.
The biggest problems I've had have been with the tracking.  Any servo
loop type function can be more difficult than it first appears and this
was no exception.

Of all the prototypes, the best was based on a simple hand scanner.
Take out the innards containing the electronics and image sensor, saw
off the original lens then mount the rest in a box with a different
lens at the front.  Fix the base of the scanner containing the rubber
wheel and encoder mechanism onto an arm so that it rides on the top of
the roll.  (You _must_ measure the paper progress linearly, forget
about encoders on the take-up spool).  Mount the sensor behind the roll
and place a high frequency fluorescent lamp (caravan light) in front of
the roll and off you go.

Now the problems.  I never found a lens with the right character-
istics to give an un-distorted image.  The biggest problem was
differential focus which degrades the effective resolution.  The other
problem is that by using an 'off the shelf' hand scanner, you're stuck
with standard TWAIN software.  I never found a package or scanner that
allowed you to create an image a foot wide and eighty feet long.
I could just about manage a dance roll (25 feet) in one scan, otherwise
it had to be done in sections.  (Even so this was with a resolution of
only 100 lines per inch.  Too low for serious work.)

The latest version uses 8-inch long contact sensor arrays (no lenses),
as found in the latest fax machines.  The world is full of these but
the manufacturers maintain an incredibly successful and infuriating
embargo on the supply of small quantities or any data sheets.  I found
a supply of mono pull-though scanners which used these devices and took
them apart to find out their drive requirements the hard way.  It was a
long tedious business with a happy outcome -- if anyone wants details
I'll be pleased to supply.  Being only 8-inches long, you need two of
these devices to cover the roll width.  The resulting two part images
are offset by 3/4-inch vertically and 4 inches horizontally and have to
be fitted together in software (much easier than it might sound).

The problem here is processing speed.  The devices have much higher
horizontal resolution than necessary and this slows things down.
I am currently working at a very high vertical resolution of 400 lines
per inch and it takes about an hour to scan a 30 ft roll.  Once the
mechanics are properly constructed it will almost certainly be possible
to drop to 200 lines per inch.  There is also great room for improve-
ment in the hardware interface to the arrays.  Even so, the top
theoretical speed is only about tempo 30, or 3 feet per minute.

If anyone wants more details please e-mail me.  I been through most
of the possibilities and will happily tell you why they didn't work !

The construction of these scanners is an important conservation
exercise, I wish more people would have a go.  Have you noticed how the
majority of original Ampico rolls already have broken webbing somewhere
along their length ?  A few more years and they'll be dust.

Richard Stibbons MIMIT
Cromer, Norfolk
England

 [ Are you now ready to make roll transcriptions for customers,
 [ Richard?  What would be the cost to transcribe a 100-foot roll
 [ for subsequent perforating?  Can you transcribe other formats,
 [ such as for orchestrions?  I think a market is waiting for you!
 [-- Robbie