HUGE collection of classical/traditional/other music midi files

[Matt Heumann]

Just stumbled across this site. You can download 5 files per day free or subscribe for unlimited. Midi files can be opened in many notation programs to learn by ear or customized by you as a concertina arrangement. Mostly classical, but under the index of "Specials": Popular Music are traditional Irish, Scottish, Welsh, American & Canadian.

https://www.kunstderfuge.com/midi.htm

Edited December 9, 2025 by Matt Heumann
additional text

[Michael Eskin]

Nice! I've been able to import several of these into my ABC Transcription Tools transcoding them from MIDI to ABC.

[Matt Heumann]

I suspected it wouldn't take you long to take things to the next level........Well Done!

Edited December 12, 2025 by Matt Heumann

[RAc]

On 12/11/2025 at 2:46 AM, Michael Eskin said:

I've been able to import several of these into my ABC Transcription Tools transcoding them from MIDI to ABC.

As a side note (apoologies for slight OT): To me it seems as if transcoding from MIDI to ABC seems a rather hard piece of work to do, at least judging from my attempts to do that using several different tools, both OS/homecooked and commercial. Whereas the opposite (trivially) is very straightforward, the quality of the resulting ABC ranges from completely unrelated gibberish to close enough.

 

Do you confirm that observation, or die I just pick all the useless tools for my tests?

 

Thanks!

[Michael Eskin]

It's incredibly fraught with peril.

 

To accomplish MIDI to ABC transcoding, I'm using a third-party open-source MIDI to MusicXML transcoder running in Python as well as a third-party open-sourced JavaScript MusicXML to ABC converter (with my own enhancements and bug fixes) built into my tool.

 

The results can range from good to horrible to complete failure depending on how the MIDI was originally created.

 

If it was a live MIDI stream capture from an actual human playing a keyboard, it's almost certainly going to result in garbage.

 

If it was machine generated and properly quantized, for example a MIDI export from MuseScore or my own tool, generally you can get surprisingly good results that can be used as at least a starting point for a piece.

 

I was able to bring in some of the J.S. Bach sonatas from that site successfully before I ran out of my free daily downloads.  Others were completely useless.

 

 

 

 

Edited December 12, 2025 by Michael Eskin

[Roger Hare]

3 hours ago, Michael Eskin said:

It's incredibly fraught with peril.   

...

If it was machine generated and properly quantized, for example a MIDI export from MuseScore or my own tool, generally you can get surprisingly good results that can be used as at least a starting point for a piece.

I can confirm that, at least empirically. I've been looking at this just a couple of minutes ago...

 

Using EasyABC, I created a MIDI file for a transcription of a simple version of Scott Joplin's 'The Entertainer'. When I imported it back into EasyABC, the ABC code I got was pretty much unusable, and the playback sounded a bit 'odd'.

 

Importing that ABC code into 'The Tool' gave me some ABC code which was certainly usable, and the playback was pretty close to the original too.

 

Trying this with a couple of random selections of human beans playing actual instruments (from the website flagged in MH's OP) did generate some 'orribly complicated ABC code (usable as a starting point for further work), though it played back OK.

 

I'll certainly be looking at this in more detail...

Edited December 12, 2025 by Roger Hare

[Michael Eskin]

Here's an example of one MIDI from that collection that worked just fine when imported into my tool:

 

https://michaeleskin.com/abctools/abctools.html?lzw=BoLgjAUAKiBSB0BleACAQgQwMYAsVQHcB7FABQwCcAXFASQDsA3AU3qoEsj6BnFAZhQQApEO5YiFZijAoATBAAy4APQAWCAEUVqgLwBOeQFkQfZQA4IAaRAARYdyIBXegBMAZlxoBbMRObDRKkkqXAAbDG4aIMd-ADVwFCDmACNQqXovHQAiAAlKAAdudlwJFyyUbgzsnMLcMoDDWhtaFHyKIgBzCgwvFAA2CHjZFGSI3ircgqKSijKKiZqxHHqRRubW9q6e-sHwCBQbNQBRNQA9ADFjtQBxM8vVFAAfFBPVQ9UAQRu7q4fni7U7y+aGUTxQpywahcZ0hXzBEKhamYMLObkRf3BkNUINUyNUCNU0OxajBX3xt1xKPxaMJJP+WJxeIJRJxT32ZNOFIJOKJBLBOI5aM5oOeXwAfCCQQBaflqMkoD5ggBeKGRYIwyglygJAH08WCmTSOqi1BrVMaMacaXjksoQWyUCqAbKxTqQTocXznkTDUi1BarXTVei8bzrUHeVjI37aZaGSi3SL0JqEQiZc948T8fdhapE399p6o1TQ0HuSSOumUB1TLIq8axScAeSg2Si2oeVTZZ8SVaq25a-3NSDSwXMR2qZ38VjuxzbidSRPp0voxiBcoADyB3OLls92PS2ebr5elA+n63RcghEO6FihGj+HFymx8sYslc+6vT9B15A77Zr88L3P+9r0uiCJwt6yjQocYLQrBIr7AhyhwdBiHwTGgYvgGNIGuito3tBYrIsiVajhhzy3GKHwADR2vR+IAMJBu8rzNt+Pxjn+m7XixGLvKc-HvBSJxVjxDw2FWcEqo8AC6uzyI6tbKipzw2IxRyMecmmMdc2mMWCWmAoxdE3AZxyGc8OkmXK9F0fRpL2c5jmis5ygOUmZnebRrnsu5aCuSgLHvPxElgvxgVqMJVxGe5qjGaoNnklZCqMVFqiRYxoVBhlZkJcoxnKugQUnK8yWcoxbpWfsHqMTFmW-O8EUdtpyg2UZqF2mo+Zgh6ygsfmO7PIm-X7joFL3J1bxNWoX6sdFgK-JcYI2VFHWPPsGmgoFtGqSSu3RVViV9YlyX6XNpl6alyUnXRVZ0YOTmap5h2Zal20JRZX3mZdv7ZfRW4XSdm0oBVF2fIxlV-T9AnfYln1MelvlBkjdlXe1QVo4jum2RiaPYij7140VzxaS5R0kzd112b9hO0xiEPnd9+UQw6t3wzTSWpQjGOeftwzPEqakHLj3OWXTyVxXjrOc0GyWfflyUZWz1li0rXMg1t9UY+L9OqKrYPq99EMg6LlNvPRpNG8oTO64bHMwxDZlOfRb2eboBNva75mGRp0n0WKaPu0TDr6QZ10R+jvvy0T5OXZH0cPSjou0QHjmyQpQA&format=noten&ssp=10&name=bach-js_two-part_inventions_3_cicking-archive&play=1

[Roger Hare]

Oh, blimey! I didn't mean to imply that the ABC code transcoded from the MIDI was always 'dodgy', I was picking up on your (ie: ME's) 'fraught with peril' remark. Plenty of tunes seem to work just fine.

 

It's certainly far better than anything else I've seen/tried.

 

Mea culpa! Sorry about that!!!

Edited December 13, 2025 by Roger Hare

[RAc]

No need to apologize for anything, Roger, I believe we all agree now that MIDI to ABC is a technically challenging process whose success is feeble and depends on a lot of factors, similar to music OCR (I still haven't found a tool that can reliably convert photographed let alone handwritten scores with reasonable quality. Lots of playground for AI, I suppose).

 

That said, I am surprised (to come back to the original topic) that many digital music archives still center around MIDI files. The ABC format, in spite of being ancient and originally primitive, appears to be superior to MusicXML as well as MIDI (any counter arguments are much appreciated), so would it not be a natural choice to use ABC files natively in those archives? 

Edited December 13, 2025 by RAc

[Michael Eskin]

I think MusicXML really is the superior option for multi-voice orchestral music. 
 

While I can't provide a comprehensive list of specific cases, it seems to me in my experience that MusicXML can represent symbols and other notation constructs not easily possible in ABC, for example, a bit of an outlier, representation of microtonal music (something I came up with a private but non-optimal workaround for in my tool).
 

Edited December 13, 2025 by Michael Eskin

[Roger Hare]

On 12/12/2025 at 6:18 PM, Michael Eskin said:

It's incredibly fraught with peril.

...

If it was machine generated and properly quantized, for example a MIDI export from MuseScore or my own tool, generally you can get surprisingly good results that can be used as at least a starting point for a piece...

Again, slightly off-topic in respect of the initial subject of the thread (apologies to the OP), but I 'investigated' this 'machine generated' business a little further (using EasyABC as a substitute for MuseScore which I don't have on my machines), I 'discovered' that if you:

 

1) remove accompaniment chords from the tune

2) remove playing order instructions (eg: P:A2ABA) from the tune

 

before generating a MIDI file, you can produce a short, relatively uncomplicated file.

 

When that is presented to 'The Tool' it generates ABC code which is indeed surprisingly good.

 

Just stick the chords and the playing order back into the tool-generated ABC code, and you're off...

 

I tried this with the same version of 'The Entertainer' which I mentioned earlier. It would still need some work to tidy things up, but hey, it works!

 

All obvious maybe, but worth pointing out? I'm sure there are far smarter folks here than me, who will be able to work out a couple of other 'wrinkles' one could use to improve this process still further...

Edited December 14, 2025 by Roger Hare

[RAc]

That may be true, Roger, but is that exercise not somewhat academic? What you normally get is the bare MIDI file itself with no information about nor control over its origin. And if you did, you probably could generate a perfect ABC right ftom the source instead oft the MIDI detour, vor am I mistaken?

 

Habe a nice Sunday, ye all! 

[Roger Hare]

2 hours ago, RAc said:

That may be true, Roger, but is that exercise not somewhat academic? What you normally get is the bare MIDI file itself with no information about nor control over its origin. And if you did, you probably could generate a perfect ABC right ftom the source instead oft the MIDI detour, vor am I mistaken?

 

Habe a nice Sunday, ye all! 

Yup! That's really the (unstated) point I was making - it was an 'exercise for the student', not much more (and after all, I was a sort of 'half-assed academic' when I was earning my daily crust🙂). As you say, in the restricted circumstances I was trying out, there's a reasonable chance of generating a 'perfect' ABC file, but in practice, we have no control over any 'real' MIDI files we encounter, and have to take any 'hit' associated with trying this stuff with such files. We get what we get - and what we've got now is very much better than what we had a couple of weeks ago...

 

It's fun though, and 'interesting', and it keeps me off the streets, and out of the public houses...🙂

Edited December 14, 2025 by Roger Hare