My Roll Scanning Project
By Mark Fontana
A quick re-introduction: I'm one of the people Jody mentioned working on an optical scanning system for 88-note rolls. I'm using a Standard spoolbox mounted in a 1920's radio cabinet. The tracker bar is equipped with IR phototransistors connected to four Intel 8255 digital I/O interfaces on a 386/40 PC running custom software. A single 150W floodlight (running on DC) serves as the exciter lamp.
Walter wrote:
> Subject: Archive of Piano Roll Material (Walter Tenten)
>
> One other property is the "algorithm" that has to be used in some cases
> to convert the input above to the notes.
>
> This includes the common problem of removing the (stability) bridges
> between adjacent holes in one track, that are ignored by pneumatic
> sensing but mostly not by optical methods.
This problem is easily solved by processing the data in a buffer, such that you can look ahead in the future when processing a given note.
The algorithm is simple. For a given sample of the tracker bar:
- if the note is not changing state, do nothing
- if the note IS changing state:
- if it's turning ON, always turn it on
- if it's turning OFF, look ahead in the future and see if that
note will be turning right back on again. If it is, skip the
note-off. Otherwise, leave the note on.
Note that this will not affect the starting or ending times of a note; only the unwanted strobing caused by the bridges will be eliminated. The critical part, of course, is determining how many samples (how long) to look ahead in the buffer. If you scan the roll at a fixed rate, it's not hard to come up with a value that gives good results.
Scanning a roll at 400 samples/sec, 7 feet per minute (tempo ~70), I've found that looking ahead by up to 20 samples (50 ms) works okay. It's even possible to do this processing in real time, 400 times per second, in between samples.
However, I'll admit I haven't investigated this all that thoroughly yet; I'm still having mechanical difficulties with my scanner. The electronics and software are ready to go, but I need to install a new motor and devise an optical auto tracking system (I have the servo for this, but no microcontroller and no idea how to begin. Tracking is complicated with old rolls that have rough edges; there needs to be a little delay before the adjustment kicks in. But if you wait too long, the roll mistracks!)
I think it's imperative that we have a system that can scan rolls in REAL TIME. There is so much material available to convert that having a slow system will become very restrictive. I'm talking about data acquisition only... editing the rolls once they've been scanned is a different matter.
By the way, if anyone wants to offer assistance with the mechanical aspects of my project, I'm very open to collaboration. I was thinking of seeing if local band organ wizard Ken Smith might want to help. He has a large collection of older 88-note rolls, too. (Check out the picture of his "Der Bebe Ruth" band organ on page 329 of the AMICA Bulletin, Vol. 31, No. 6, Nov/Dec 1994. I've heard this instrument, and it's marvelous! Ken Smith is also a union projectionist and a collector of antique motion picture equipment. He's got a "museum" behind his home with at least a dozen projectors (dating from about 1910-1960), and even a 35mm dual projector screening room in his home with carbon arc illumination!)
Mark Fontana |
(Message sent Sun 26 Nov 1995, 03:33:30 GMT, from time zone GMT-0500.) |
|
|
Mechanical
Music Digest™ Archives
