Digitizing Player Piano Rolls

What do you do with 100 player piano rolls but no player piano? You come up with a way to digitize the information for MIDI playback. The rolls have 90 columns worth of holes, 88 for the keys and two more for pedals. Voids in the paper cause a note or pedal to be played, so an optical sensor can be used to transform the analog data into digital information. Simple enough, you’ll just need 90 sensors. But this brings up quite an alignment issue. The solution is to use fiber optic cable to position the IR light source in a hand-made 0.2″ spaced jig. At least the spacing meshes nicely with standard 0.1″ protoboard, which is what was used for mounting the sensors.

[Thanks Mike]

28 thoughts on “Digitizing Player Piano Rolls

  1. Neat! I think hes just about ready to Have intallation #3 of this build up, I know most of the mechanics are done, its now getting close to the digital side of things.

    If anyone in the Albuquerque area is interested in such things Quelab is Albuquerque’s Hackerspace.

  2. Thought: might it have been easier (though way less cool) to use a scanner or webcam to capture the rolls in transit, and figure out the rest in software? Similar to to all those automatic guitar hero projects?

  3. >so an optical sensor can be used to transform the analog data into digital information.

    There’s nothing analog about this data – it’s still digital data, even though it’s 50 years old and embedded on a paper roll.

  4. I thought this sounded like the hard way to do it. plus wont you get light leak?

    Not that the creator should feel discouraged. in fact the level of workmanship is really impressive. I still enjoy reading these kinds of posts.

  5. There are perhaps easier ways to do this, but I think it’s refreshing to see a different take on the problem.

    @Andrew:

    It’s two-dimensional data, digital in one direction (i.e. discrete positions for each key or pedal), but analog in the other (i.e. the length of one of those holes could be any value).

  6. Not to detract from the awesome looking machine, but some players let you “feather” the holes to adjust the playback, beyond having soft/sustain peddle tracks. This seems to only store hole/no hole instead of light intensity or whole width that a scanner based method would do.

  7. This reminds me of what happened when the Smithsonian discovered that they had in their collection a remarkable collection of piano rolls cut by such composers as Rachmaninoff, and Gershwin. They needed the piano that cut them to even play them back. It turns out of course that they had that as well. It was a simple matter to assemble a chamber orchestra of suitable instruments in their collection to play back the compositions.

    Of course that was an extreme case that allows for the availability of everything.

  8. I should also mention that the basic piano roll is about the size of a standard punched card. A computer over in Britain who used a similar method for running an Ammonia making plant, who was called ATLAS was re-purposed to run a sort of music making system not too long ago.

  9. While many are quick to point out that there are several other easier ways to simply make an archive of the music on these rolls. The potential I see with this method would be the ability to not only make an archive of the rolls but to also play them back in real time via the MIDI interface.

    Now if that is indeed the direction the author is going in well, IMO, major kudos to them because that’s pretty freaking cool.

  10. IBeHoey, Indeed That is part of it, We want it to be a working player, When it was found it was just the gutted paper transport, which in its self is awesome. it has a amazing self centering method that i really wish they had added to tractor feed and plotter printers! Looking at the old Lead tube vacuum lines weresmashed and cut, and it looked like a neat bit of history sestined for the junkyard.
    When this project is done, it will allow us to digitize the tracks, but that could have been done lots of ways as others have posted, this will allow us to live play them any number of ways, both in sound and in lights or other methods.

    All of which are more fun than just having a copy of a old time midi file on a computer.

  11. An Aeolian spool box no doubt, the finest. Tracking is disconnected, but should be hooked up with a servo to the two fingers at each side of the tracker bar. This keeps the roll centered on track. Players never encoded the soft pedal only the damper pedal. The exception of course were the reproducers like Aeolians DuoArt with 4 bits for each half of the keyboard that controlled volume. This was binary weighted 1-2-4-8 in 1913! Midi only uses twice the bits but for each note. These players were fully automatic, incl reroll and shutoff. Whats missing from a roll conversion to midi, is the absolute joy of playing with your feet the volume of your own expression. This should be done in midi with a control in real time or edited into the work. A player running on a suction source is as boring as hearing recordings that have been compressed or peed on by mp3. Thanks for the links. I will follow them up, as I rebuild players and want to hack a roll perforator to preserve old rolls and bring new music to this old digital realm. Acid paper is taking it’s toll. At it’s peak the player was in 85% of pianos sold and the home entertainment system of choice.

  12. “All of which are more fun than just having a copy of a old time midi file on a computer.”

    Couldn’t agree more and I’m very happy to hear that you are going in that direction. Very, very cool. I’ve always been fascinated by the old player pianos and really hope to own one someday. I just might have to end up following in your footsteps and building one for myself.

    You sir have inspired me, and for that I thank you.

  13. Very happy to see this – my grandmother still has a working Pianola and a library of 30+ rolls in her house (many christmas carol rolls, even the entire musical score of CATS)

  14. Reminds me of waking up to “O What a Beautiful Morning” when I was a kid, when my parents had a player at the bottom of the stairs.

    I fondly recall switching the piano to the twangy setting (pulling a small plunger under the keyboard) so it sounded like something from an old western. Dad didn’t like it when I did that :p

  15. @heegemcgee
    In all likelihood that was the second time this had happened. See my first comment. But the first time they had done what is going on here. I remember listening to the album on WQXR-FM, the new NPR station in NYC last year.

  16. hackaday, is there a typo here?

    The information on the piano roll is already “digital” and “binary.” Scanning it optically is actually a form of conversion, not ‘digitisation.’

  17. Many years ago I made a piano-roll (and band-organ book) reader from parts of an old FAX machine. The FAX machine had a line-scan CCD camera, analog compensation and quantization for the video, and several channels of 1-bit memory.

    Metal templates were made for each of the different kinds tracker bars to be supported. Care was taken to make the templates with exactly the same hole spacing as a tracker bar, but with holes about half the diameter of the tracker bar hole width.

    When the template was inserted between the light source (a florescent tube powered with DC) and the line-scan camera, the resulting digitized `video’ was recorded in on of the memories, that I called the template memory. This was `training mode’.

    Later, in `use mode’ the roll or book was fed through the machine, the system was effectively sensitive only to the light that passed through the centers of the holes in the book.

    Each camera scan of the roll or book traversed from one side, and as it went the memory address incremented at the camera pixel rate to 4096 across the width of the book. Each of the template `images’ consisted of several consecutive bits in the memory.

    In `use mode’, the camera video was recorded in one of the other memories. The digitized camera video was also compared with the recorded video from the previous scan (stored in the the other data memory. The comparison was done using a 2-line to one of four decoder to easily detect Note-ON and Note-OFF. The comparison was gated with the output of the template memory, causing the sampling to be done near the center of each hole (width-wise).

    At the end of each camera scan, the two data memories were swapped.

    Also at the beginning of each camera scan, a counter was initialized, and then counted each template memory bit as the camera scan progressed. When the counter was initialized with the proper value, the counter gave the MIDI note number for each row of the roll or book directly. Thus each Note-ON and Note-OFF would have the proper note number.

    Strictly speaking, the longitudinal sense area should be equal in length to the length of the hole on the tracker bar. With this refinement, the MIDI codes would more accurately mimic the action of the player piano. (Otherwise, the rapid Note-ON and Note-OFF sequenced that would result from sustained notes on the roll would be objectionable, and would have to be adjusted in post-processing.)

    The piano rolls I recorded were for Ampico, one of the varieties of `reproducing piano’. Several rows near each end of the tracker bar are used to control the `expression mechanism’ of the piano. To simply play the rolls (i.e., without expression), the signals from the `expression rows’ rows (i.e., certain of the MIDI notes) should not be forwarded to a MIDI synthesizer.

    A more elaborate system would use the expression signals and some simple analog circuitry to model the behavior of the pneumatic servos in an Ampico (eight levels of expression plus, fast and slow crescendo, decrescendo and faster discharge). The result, digitized, would be used to set the MIDI `pressure’ value for each MIDI Note-ON.

    Incidentally, an Ampico has only two expression `channels’, one for `bass’, the other for ‘treble’, divided a few notes above middle `CV’. MIDI has individual `pressure’ control for each note.

    (I’ve left out some details of the timing, and also a description of the `tracking’ mechanism to keep the roll in the right position as it is scanned. In a player piano, the roll is movable laterally by a pneumatic servo that tracks the edge of the mving roll near the tracker bar. In my system, the `tracking’ was done digitally, by sensing the edge of the roll with the camera. The `carraige’ holding the tensioned roll or book was movable laterally by a small, reversable DC motor controlled by the sensed video. Using one more memory, it would be easy to derive a atracking error signal from each note row.)

  18. Does anyone think this system could be adapted to play the paper roll’s used on Wurlitzer Band organs? I have several rolls that were used on a style 153 Wurlitzer Band Organ used on an amusement park carousel. The Organ was sold at auction, but I still own the original roll’s and one of the roll reels that was used as a spare for quick change of the music roll’s. Chuck L. Connecticut.

Leave a Reply to brainzillaCancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.