You could take a "pure" mapping of just the notes and relative durations... but then you'd wind up with what would essentially sound like a weird robotic "player piano" version of the music at strange speeds that would be, well... pretty bad.
Back in 1993 Yamaha built a device that could read piano rolls personally “recorded” by Gershwin by hand (he would hammer them out for money in his spare time), and then played back on a modern player piano. The resulting album is one of my all-time favorite albums.
Yamaha still sells these “scans” (on 3.5” floppy!) for their player pianos, and they could certainly be reverse engineered, but I imagine they are considered derivative works and covered by their own copyright.
https://apnews.com/fb3f8c3bc3305506f57120df0755f9d8 https://youtu.be/_kIpr6nSvjI
But I'll point out that the YouTube performance you linked to isn't a player piano at all -- it's an actual recording, which Gershwin overdubbed to be two pianos.
While player pianos that could incorporate dynamics existed, I'm not sure they were common, and I'm also not clear if there were any that incorporated playtime "per note" dynamics as opposed to generally "overall" dynamics that were added in via a separate mechanism.
But nevertheless yes -- there did exist some player pianos that had the capability of being more expressive than the normal single-volume ones!
This[0] is a reproducing piano recording Ravel made in the 20s and was recorded from a Duo-Art piano in the 60s. Hearing Ravel play his own stuff a few years ago radically changed my idea of how Ravel's music should sound. (Pianist, I'd been playing Ravel for a couple of decades before that) This[2] has Prokofiev playing his own music.
"Many famous composers from around the world played their own works for the reproducing piano: Edvard Grieg, 1906 in Leipzig; Alexander Scriabin, 1910 in Moscow; Gabriel Fauré, c. 1913 in Paris; Nikolai Medtner, c. 1925 in New York."[1]
"This video is a practical demonstration and overview of how the Aeolian Duo-Art pneumatic reproducing player piano works, including how the system plays expressively by controlling the loudness of notes played with perforations on the roll."[3]
[0] https://archive.org/details/RavelPlaysRavel
[1] https://www.allclassical.org/player-piano-rolls-listening-to...
This YouTube rip sounds a little crummy to me, though, I’d suggest finding Gershwin Plays Gershwin on Spotify or your streaming service of choice...
The YouTube link you referenced is an actual recording -- the notes are all distinctly different dynamics, and there's plenty of background hiss. It's a real recording, full of life, on a real piano, and genuinely overdubbed. It's not "a little crummy" -- to the contrary, it's an invaluable historical record.
On the other hand, the "Gershwin Plays Gershwin" album is a genuine player piano, but the Rhapsody in Blue is totally different -- it has zero dynamics, everything is exactly the same volume. [1] It's interesting to hear, for sure -- but more as a historical curiosity, since no pianist would ever perform everything at the same volume unless limited technologically. Your original link is Gershwin's full interpretation. In contrast, his player piano version is nothing like what he would have performed live, but rather adapted to the technological limitations.
[1] https://open.spotify.com/track/2XSBXz4uDvx1PQPYJWQpcK?si=EO3...
The sleeve notes tell of how modern scanning allowed a more complete reproduction of the notes played than contemporary players allowed, which by all accounts were already pretty good.
And what if my program is parametrized by an ML model of unknown training data?
Then that program will likely produce non-copyrighted output for non-copyrighted input. But it's also likely going to sound pretty bad.
> And what if my program is parametrized by an ML model of unknown training data?
Then its output could be considered a derivative work of every work in that training corpus, and anyone whose creative work went into that training data will have a copyright claim on the results.
Depends how tech savvy the judge is. An ML model is not really like a copy or reproduction, it's much more like a person's brain in that it aims to take general patterns from various sources.
Instruments where notes/hits are isolated from other notes/hits like a piano or drumkit work wonderfully with sample libraries. But things like a guitar, violin, or trombone are still challenges. Even the orchestral libraries that are hundreds of gigs and thousands of dollars don't get it perfectly on their own. You definitely can make recordings that trick a trained musician, but not without tons of effort customizing your midi transcription to that specific sample library and you may have to compromise and avoid certain techniques that still don't translate to sampling well.
This is just talking about trained musicians though. Taking a midi transcription of something, spending 10 minutes adjusting velocity, and then loading up some EastWest samples would probably fool the average person.
Also quick edit but many people would consider something using samples of recorded instruments to still be a midi synthesizer because you're synthesizing a performance from samples, as opposed to recording entire phrases or even longer sections with a real musician and then arranging samples to create a recording. Not everyone would agree, but I'd guess the original commenter also had sample software in mind along with more traditional electronic synthesizers.
While we're on it, drums aren't totally isolated either, they affect themselves. A snare roll is slightly different than just playing many snare samples rapidly because of the chains on the drum, and heavy cymbals also take some time to build up momentum so the first hit and tenth hit will sound different if it has not settled. These are all very subtle though compared to something like a guitar or violin or trombone where transitioning from one note sample to another sample digitally can be very obvious.
Disclaimer: i by no means have any expertise on the topic and am just as curious about figuring out how it actually works as the parent comment.
Since you're talking about the Raspberry Pi, maybe you're looking for a hardware. In this case, you can get MIDI piano keyboards that plug into the computer over USB (or can be used standalone). These can both send MIDI data to the computer and also receive it and play it with their internal synthesizer. (If you go really expensive, you can get actual grand pianos with sensors and actuators under each key, along with a subscription service that gives you access to recordings by professional piano players, replayable at your whim with exact dynamics so it's like they're in your living room.)
You can also build your own hardware and hook it up with a microcontroller; I've had good experiences with the Teensy, which is similar to an Arduino but can present itself to the computer as a USB MIDI device so you can build whatever creative/crazy instruments you want.
Should you come across a device with an actual MIDI port (a big round DIN socket with 5 pins), it's just a serial connection; you can get adaptors to convert them to USB.
However, unless you transcribed the original sheet music to MIDI yourself, the file may constitute an “original work” for which there is a separate copyright[0], in which case you'd still need an appropriate license for that. You'd also want to make sure that the sound fonts the synthesizer is using are OK as well.
Now that the original work is in the public domain, though, I would expect that sooner or later there will be a CC-licensed version of it on Wikimedia Commons or the YouTube Audio Library, and you may want to just wait for that.
[0]: https://meta.wikimedia.org/wiki/Wikilegal/MIDI_Files