Piano Rolls

I’m working on the discography of Fats Waller which has led me to thinking about piano rolls. Waller recorded several of these early in his career, under his real name Thomas Waller. (There were also piano rolls sold as “Played by Fats Waller” but that were in fact played by J. Lawrence Cook in Waller’s style.

It seems reasonable to me to treat the piano roll itself and later LP/CD releases “from the original piano rolls” as the same recording, as in this merge.

It also seems reasonable to credit Waller with “piano” on the recording. Piano rolls were often manipulated after the initial recording, but that’s not qualitatively different than later techniques like editing and overdubbing.

Then we come to oddities like Fats at the Organ. Waller was an accomplished organist, but this release was made from piano rolls used to play an organ. I’m not sure how to credit that, nor how to relate those recordings to the originals.


I don’t know that I’ve got an answer, but another similar example. MIDI files are kind of in the same boat, as piano rolls and MIDI files are more like sheet music, which is closer to a work than a recording in my mind. that said, there should be a way to fit these into our database somehow, but I don’t have any particular ideas… maybe a release attached to works and not recordings? seems odd…


I don’t see them as especially comparable to sheet music. No two performances of the same sheet music are going to be exactly the same; every artist will interpret it in their own style. In the case of the piano roll, the artist’s style is captured in the roll.

“Piano roll” is already a supported release format, and has been for a long time, so I don’t think there should be any controversy about that.


yeah, but a piano roll played on two different pianos wouldn’t necessarily sound exactly the same? I don’t know, I was more referring to the fact that like MIDI and sheet music, it’s basically instructions for creating music. definitely more precise than sheet music though


I agree that two recordings made from the same piano roll aren’t the same recording. They are definitely closely related, but there are so many different variables, both the instrument itself and the way they are recorded. I think MIDI is quite similar, because the same MIDI file will sound different with a different soundfont.

I guess both MIDI files and piano rolls can be seen as sheet music + timings.


Each piano has its sound, so the recordings are not the same, IMO, even if they follow the same instructions (piano roll).


I feel that an audio recording from a piano roll is almost like a live performance of a work. A different sound from the different machinery. Like a band picking up different guitars \ drums to use in a different building with different acoustics.

I’m no expert but feel like they should be kept separate. The Piano is the performer.


It’s hard to tell if a piano roll should constitute a recording or not without knowing the nature of the piano roll in question. The Wikipedia article on piano rolls explains that there are diffent types of piano rolls:

  • created through stenciling without performance (akin to sheet music that can then be performed by a “player pianist” making essential creative choices using the note and timing data as input)
  • created through recording basic note and timing data of an actual performance (but where any further performance of the piano roll requires intervention by an operator who would make choices about the dynamics of that performance)
  • created through recording note, timing and dynamics data of a performance (that can then be played back with little to no intervention by an operator, fairly closely to the style of the original performer, and varying mostly in choice of speed and reproducing instrument)

While I would consider the first type very similar to sheet music (and thus not a recording in itself in a strict sense), the latter two kinds of piano rolls certainly are recordings as far as I’m concerned.

No matter what kind of piano roll, though, I agree that individual recorded performances of piano rolls should be regarded as separate recordings for the same reasons stated repeatedly in this thread.

For the piano roll as a medium itself, it seems reasonable that Waller should be credited with “piano”. For subsequent recordings of piano rolls, I guess Waller should be credited with “piano” as well, and depeding on the input of the pianola player (who can add further effects), that person could be added as a further recording artist.

Some examples…

For “Gershwin Plays Gershwin”, the pianola operator Artis Wodehouse (according to Wikipedia) isn’t mentioned on MB as a recording artist or anywhere in the artist credits. I like that the recordings associated with this release consistently include annotations that point out they’re based on piano rolls and, if available, when/where they were recorded.

For the “Piano Concerto” release, recordings by pianola player Rex Lawson using Edvard Grieg’s piano rolls feature both Grieg as well as Lawson as recording artists. Perhaps Lawson added more in terms of creative decisions than Wodehouse on the Gershwin release?


I’ve cancelled my open piano roll related merges based on the comments here.

However, it seems to me a bit inconsistent with how we handle mastering. Every digital mastering of an analog source is going to sound a little different, but for MB purposes we say “close enough”, we’re more focused on the performance itself.

Interesting thoughts from @nadim regarding the pianola operator. Credited or not, they could potentially have a significant impact on the final sound. (But so too can the folks applying noise reduction to a 78 transfer…which, again, we don’t generally keep separate.)


I don’t agree with that at all. The performer is the performer!

But…your comment about live performance reminds me that the guidelines do say that “recordings of the same performance from different sources” should be separate recordings. That seems a better analogy here; a soundboard vs audience recording of a live show are distinct even though they capture the same peformance.


You bring up a good point that I didn’t catch when looking at piano roll recordings last night. Looking at the piano roll George Gershwin recorded for “An American in Paris, Part 1”, it turns out that a later recording of audio produced by playing that piano roll on a pianola is linked to the original piano roll recording with the “remaster” relationship (which is deprecated, creating the issue that piano rolls and audio recordings based on them may be inadvertently merged in the future - which would be unnecessary and undesired if there is an actual consensus that they should remain different recordings). I think a remaster relationship is technically incorrect, because recordings of piano rolls played on some instrument generate completely new source audio, so they can’t be remasters in MB’s audio-centric definition of the term.

How would we associate piano rolls as recordings and audio recordings based on piano rolls? MB doesn’t have a recording relationship that seems fully appropriate. Perhaps the “remix” relationship: a new mix has been created from previously existing audio elements. Or the “sample” relationship: an element of a track was used in another track, perhaps heavily processed. But then we run into the problem that a piano roll is a recording of performer input, not the actual audio that resulted while the performer played on a recording piano - and MB’s definition of a recording is very much centered around “audio”, regardless of whether intentionally or unintentionally: most media formats we’re dealing with on MB transport audio signals, after all.

There is an MB term that looks appropriate but doesn’t seem to be used in practice. The MB terminology page defines an “interpolation” as follows: “A type of sample in which the melody of a source is copied, rather than the original audio being used.” That would fit rather nicely for audio recordings using piano rolls! Unfortunately, there is no recording relationship based on that definition, and there’s little to be found searching the rest of the documentation for “interpolation”. I’m not sure how I would go about a feature request for such a new recording relationship - but the first issue would be seeing if anyone else agrees an “interpolation” would be appropriate in this case.


Instead of trying to shoehorn an existing relationship to accommodate piano rolls, wouldn’t it be better to create a new relationship “is recorded from the same piano roll / midi file”? Most relationships have a direction (X is a Y of Z) but that is of course not necessary.


A direct relation between the piano roll itself and the recording based on it would be better. It’s (generally) going to be a 1 to 1 relationship, which “recorded from the same piano roll” would not be.

But wouldn’t you need a new object type for that, to sit between work and recording? Because I doubt that is going to happen.

“Piano roll” is already a supported media type, so the roll itself is a recording. This would just be a new recording-recording relationship.

1 Like

Yup, I don’t like repurposing existing relationship types (or extending their purpose) either. That’s why I started thinking about the “interpolation” that is in MB terminology but seemingly nowhere else on this site - however, looking it up meanwhile, that turns out to be more of a musical citation, not a sample, but not quite a cover version.

If we’re working on a proposal for a new recording relationship type, I’d like to make it agnostic of storage formats. Perhaps “is a playback recording of / has playback recording”?

(Funnily enough, I just had the thought that I phrased my suggestion so generally that it could possibly also be applied to pre-recorded music played over the house PA at concerts if a sufficient part of an original recording lands on a live recording. I’ll be over here now… :wink: )

1 Like

Other ideas…

  • “is a replay of / has replay”
  • “is an interpretation of / has interpretation”

I like the term “interpretation” because it suggests, just as with a classical score, that a basis of music notation is used and interpreted (by e.g. a pianola player) with a certain amount of creative freedom as to instrument and other choices of expression. In fact, looking at web sites that describe piano rolls and pianola/player pianos, that word tends to be used frequently.

I think “interpretation of” is a bit overly broad, and maybe doesn’t fit the MIDI use case well.

  • “is audio expression of / has audio expression”?

The thing with “interpretation” is that it does come up quite often on web sites about piano rolls and pianolas and other reproduction devices. I think it would fit MIDI decently as well, because the choice of sound font (and possibly some additional effects) is an interpretation of the source music data.

My thought about it being overly broad is that it could be applied to a lot of non-mechanical “interpretations”…like, artist B’s recording of work X is based on/inspired by artist A’s recording so that could be considered an interpretation.

1 Like