It’s dead Jim.

I assume we hugged it to death.

Super cool. Clicked on a title with the MIDI indicator and a MIDI file (.mid) downloaded. Came up in Garage Band and sounded nice.

If you find a duplicate it often isn't. They often cut a bunch of rolls and then threw the master. If the roll proved popular they made a new master which would be slightly difierent but have the same catalog number. Tracing these 'editions' is often part of the fun.

If you want to play these in VLC you need a SoundFont (.sf2) file. There's a good list of SoundFont files here [1]. This VLC wiki page [2] explains how to configure VLC to use the SoundFont.

[1]: https://github.com/FluidSynth/fluidsynth/wiki/SoundFont

[2]: https://wiki.videolan.org/Midi/

(I'm posting this because the vlc wiki is stale and sent me down a pointless rabbit hole on fluidsynth's old sourceforge site. I'd rather update the wiki. It tells me I need to create an account. When I try it tells me I don't have permission.)

This used to be one of my main hobbies, I listened to thousands of these and I am super grateful to the people scanning and hosting these collections.

Some software I wrote for piano roll analysis and transcription:

- Unroll: https://zulko.github.io/unroll-online/ - upload a piano roll midi file and have it quantized and converted to lilypond sheet music. More about the process in this blog: https://zulko.github.io/blog/2014/02/12/transcribing-piano-r...

- Pianola: https://zulko.github.io/pianola/ - upload a piano roll midi file, and it plays with the piano roll and keyboard animation (you can zoom on some parts, slow down etc).

Some transcriptions made with these tools:

- Hindustan: https://github.com/Zulko/sheet-music--hindustan

- Gershwin - Sweet and Lowdown: https://github.com/Zulko/sheet-music--Gershwin-sweet-and-low...

- Gershwin - Limehouse Nights: https://github.com/Zulko/-sheet-music--Gerhswin-Limehouse-Ni...

An interesting prospective project for a technically minded musician would be would be to find an automated way to "correct" the surviving corpus of Welte-Mignon[1] recordings. They were designed to capture the small nuances of performances (such as dynamics), and a large number of historically important musicians made recorded performances in this medium before the era of sound recording. In my strongly-held opinion, the rolls were marked in an uneven and imprecise way, making direct playback on anything but the original recording apparatus inaccurate. A common trait of modern renderings of these rolls as sound recordings (as found on CD or on Youtube) is an unevenness of tempo and a seeming lack of synchronization of voices (really piano keys). However, the mechanical quirks and imprecision in the recording apparatus must be regular enough to allow for a more accurate version of the performances to be reconstructed, without relying on unduly many aesthetic assumptions.

[1] https://en.wikipedia.org/wiki/Welte-Mignon

"The neverending piano": Feed these to a tokenizer, vectorize, generate ... forever.-

At the bottom of page 1:

>Because of U.S. copyright restrictions, only songs published in 1929 and earlier available for public download from this page.

What am I missing here? There are a list of piano rolls that have been scanned but you can only download MIDI from before 1929.

I mean, I understand copyright and public domain law in the US, but what exactly is the point of a list of things you can't share?

It is a shame that you can't access the scanned piano rolls of Conlon Nancarrow's player piano music in an archive like this. The scans do exist apparently, but are merely documented on the web (https://www.paul-sacher-stiftung.ch/en/research/1-2-Conlon-N...) without explicit download access. An incredibly comprehensive Youtube video exists however: https://www.youtube.com/watch?v=AfmVJmQKdj4