Drawing inspiration from the great Conlon Nancarrow, a pioneer in acoustic player piano manipulation well before the era of MIDI and DAWs, I find myself contemplating a unique but surely controversial idea: I wonder, what would be the outcome of creating "computer music" designed for a MIDI-compatible yet acoustic player piano? The execution involves feeding meticulously crafted MIDI data into the player piano through systems tailored for select models capable of interpreting MIDI files. The programming dictates the execution, yet the acoustic piano recording would be entirely authentic—captured in a breathing moment of time.

Adding an intriguing layer to this concept is the ability to infuse expressive nuances into MIDI data—a well-established technology and a known technique for producing convincing MIDI mock-ups.

Does anyone know if this has been done in the past? More importantly, if so, what did the results sound like? Solely listening to Nancarrow's recordings, the result appears quite artificial, yet with a subtle nuanced charm added by the true acoustic piano.Bear in mind, Nancarrow's writings were not exactly well suited for "Human Performance." Works written for true piano live performance would undoubtedly be more convincing.

You need to be a member of composersforum to add comments!

Join composersforum

Email me when people reply –


  • Hi,

    something like that was very well done by karleinz Essl, the Lexicon Sonata:


    Karlheinz Essl: Lexikon-Sonate - algorithmic music generator (1992-2020)
    Interactive realtime composition running as software on Apple Macintosh computers.
    • Close, but not quite. That is an interesting approach for that work, but I am referring to a pre-constructed set of MIDI data created in a DAW, then being fed into a player piano that plays off of MIDI data. The Sonata here only seems to be using the piano to act as an amplification and resonation device, but it is not producing the initial tones themselves. 

  • You mean something like this?

    Player piano
    A player piano (also known as a pianola) is a self-playing piano containing a pneumatic or electro-mechanical mechanism that operates the piano actio…
    • Or, to elaborate, player pianos controlled by MIDI already exists; what you're proposing is just a way to program the player piano with human-like nuances.  Which is no different from the analogous production in a DAW, really, except that instead of using samples in the output, you use actual acoustic sounds from the player piano.

      Taking this even further, what about other acoustic instruments that are MIDI-controlled?  What if there could be a mechanical violin, controlled by a robot that actually executes bowing movements to play notes, directed by a MIDI controller.  Or a flute with a blowing mechanism attached. Or a trumpet with rubber lips controlled by some mechanism for intonation.

      Taken to the logical extreme, we could imagine an entire virtual orchestra created this way, well, rather, an actual orchestra but with MIDI controllers instead of human players. In theory one *could* program all of that to produce a convincing performance. (No telling how the results would compare to a performance by real players, though!)

      At a certain point, you'd have to program all the expertise of every player in the orchestra into the MIDI controller, in order to get a good performance. At that rate, you might as well just hire real players to do it instead. :-D

      Or, since we're on the topic of AI, if we could instrument(!) each player with devices to measure the parameters of their performance over, say, 100's of performances, and feed that to an AI to "learn" how they play, then we could potentially attach trained AIs to each MIDI controller, and get an "actual" orchestra that, supposedly, plays like how the human players would.  Interesting thought...

    • You were correct in understanding that I was referring to a MIDI-controlled player piano. However, the MIDI data used is created in a DAW to attempt to capture the nuances, expressions, and rubato of a human player.

      In simpler terms, my inquiry revolves around whether anyone is aware of a recording produced in this manner. This involves feeding a well-constructed MIDI set for a piece into an acoustic player piano that reads from MIDI and then recording it in the same way one typically records an acoustic piano.

      This method could be used to acoustically record "impossible" pieces yet while capturing the charm of an acoustic instrument, or, to "cheat" on a budget and still record a work but without the fee and time of having a pianist learn and perform the work. This is obviously not ideal but makes me wonder how the result would fair.


    • How would it differ from recording a pianist's performance as MIDI, and feeding that data to the player piano to recreate the same performance as many times as you want?

      Or, for that matter, record a complex piece by splitting it into smaller, more manageable chunks, and pasting them together?  This can be done either horizontally (cut a piece into multiple sections, record each one in turn), vertically (record, say, each voice in a complex contrapuntal work, then paste them together), or both (record individual melodic fragments across multiple voices/parts, then assemble the results into a single piece).

      Or, if you want to pursue the idea of a virtually "performed" piece, you could record a pianist playing, say, each note of the 188 keys at 5 different dynamic levels across 5 articulations, then paste these fragments together like a VST does with samples. This is an extreme case, of course, which begs the question of why not just use a VST in the first place, since that would essentially be the same thing. But it's what we'd arrive at if we push this idea to its logical conclusion.

      An interesting medium is a pianist recording not individual notes, but short 2-3 note fragments that can then be pasted together to form melodies. This would avoid some of the "machine gun effect" of mechanically-generated music caused by intoning every note in a phrase identically, by encoding at least some of the pianist's nuances into the fragments.

      Also interesting is the idea of having a computer AI analyze the MIDI data of a pianist's actual performance to "learn" how he phrases certain notes when in the context of certain other notes.  An AI trained in this way would be able to, ostensibly, reproduce at least some of the player's nuances when asked to play something the pianist has never actually played himself. (Leaving aside, of course, important questions about variations in nuance caused by the pianist's higher-level understanding of the structure of the piece, interpretation quirks, etc., which the AI would be unlikely to apply in the same way to a work the pianist himself has never played, and therefore his would-be nuances in performing it would not have been present in the training data.)  Such an AI could, in theory, produce MIDI data that would drive the player piano to play as the original pianist would have.

      The question of classifying the result as "performance" or "mock-up" would then be a big fuzzy gray zone we could argue about forever without any resolution. :-D

    • Midi-controlled church organs are quite common. It's cheaper than paying an organ player.



  • At least one of Bösendorfer's pianos is wired for midi so is effectively the same thing. I think VSL take advantage of this in their offering with its gargantuan space requirement: nigh 300Gb.

  • There is a genre called "black MIDI" that is all about impossible-to-play MIDI music (usually acoustic piano sounds).  And I know I've seen videos, like this one, where they play those pieces on player pianos...


    • Oh, OK, I just realized this is a piece written for player piano by Marc-Andre Hamelin (and is considered a precursor to black MIDI!)

      Circus Galop - Wikipedia

      Circus Galop
      Circus Galop is a piece written for player pianos by Marc-André Hamelin. It was composed between the years 1991 and 1994 and it is dedicated to Beatr…
This reply was deleted.

Topics by Tags

Monthly Archives