NotaGen: Symbolic Music Generation
46 comments
·March 23, 2025jcpst
TheOtherHobbes
It's literally a list of cliches glued together. Some of them are style-defining cliches, but there is no hint of a strong original voice. And it's the individual distinctive voice of each composer that defines classical music. The musical wrapping is the means to that end, not the end itself.
The waltz sounds like elevator Chopin with all the bittersweet irony sanded off (and couple of questionable harmonies.)
It's an achievement to get this far, but it's still very uncanny valley, and (IMO) a fundamental limitation of LLMs that can't see beyond the superficialities of weightings into what the structures mean.
oersted
I was actually quite impressed by the non-human version of the Waltz before seeing the video with the musician. It's a lot like the fantastic Succession theme, it's not a generic waltz, it's a bit dark in an interesting way.
janalsncm
I am really interested in this. The current paradigm from e.g. Suno is an all or nothing finished product. Producing intermediate assets allows you to do simple things like proper mastering or swapping instruments or editing melodies etc.
lotyrin
Yeah. I want a DAW with autopilot features (assistance), not for the LLM to wholesale take away my creative input (vibe composing).
metaketa
This is our goal with https://parture.org!
titaphraz
Is your goal to break down audio to individual instrument's sheet music? That'd be nothing short of amazing.
dwallin
I agree that what ai music needs to become an industry tool is the ability to create, access and remix parts, but I think tools like Suno have more of the right idea vs tools like this. In order to be able to write intermediate parts properly, you need to be able to understand the whole and what things should sound like when put together, or when the notes are actually played by a musician. Then it’s easier to work back from there, split your tracks apart into stems, transcribe your stems into MIDI, etc.
Suno et al are moving in this direction but I honestly think development will be somewhat stunted until we get a good open source model(s), and something like control-nets.
resource0x
An impressive achievement indeed. That's exactly what one'd expect from LLM. If someone wants to see/hear what real music is, listen to this, just to get a reference point. This will make you cry. Literally.
-__---____-ZXyw
For anyone else tempted to give in to their baser instincts, don't waste your time on the links, it's two very short piano performances. They're good, but it's statistically extremely unlikely that you will shed even a single tear.
I am generally in favour of throwing shade on LLM "creativity", at least thus far, but this is an uncomfortably strained flex. Literally.
zoogeny
I think the "literally cry" thing is a zoomer thing. I watch a lot of YouTube music video channels and the "this made me cry" thing is a meme that shows up with extreme frequency. I don't think we should take it any more seriously than we do "laugh out loud" when the actual response is usually more like a quick exhale.
CuriouslyC
What is it about classical and jazz fans that makes them so pretentious?
resource0x
Please listen to the above pieces once a day (it will take only 4 minutes) for 10 days. Then, while speaking about music, you will sound pretensious to your friends. Please try. (I can send more. This will include jazz certainly. And Beatles... and others. You have to put some effort into it, it will pay off).
CuriouslyC
I've been studying music for 10 years, I'm not some 12 year old who's only ever heard pop.
Classical music isn't special superior form of art, and people who try to paint it as such tend to be elitists who want to have something to hold over other people to make them feel superior. Kind of like how people latch on to wine to try and seem sophisticated (ironically, wine snobbery has its origin with kings and chieftains of conquered territories trying to fit in with their Roman oppressors, but at least they were cognizant of why they were doing it, unlike modern wine snobs). There's plenty of metal that's far more technical and complex than typical classical music, but you don't see metalheads going around telling people to listen to "real" music.
peterburkimsher
I was expecting another song, which is more likely to elicit an emotional response.
YeGoblynQueenne
Hello fellow emotional music fan. This makes me cry, literally:
dimatura
The pop composition was nice, if somewhat cliched. (And of course the MIDI instrumentation does it no favors). The guitar solo in the middle was fun - it kind of sounded like it reverted to some of the baroque pieces in its pretraining. Reminded me of the harpischord solo in the Beatles' "In my life".
yapyap
Generative music is boring not cause it’s necessarily bad sounding music but because there’s no vision behind it, they took in old music, garbled it around in their magic music machine and made new music based on that old shit.
As long as there’s music made by real people this AI music stuff will only be interesting to the very uninspired, the people that will hype up anything that’s AI and commercial projects who don’t have to pay royalties for the music in their advertisements, elevator, phone waiting music and whatever else they need a quick tune for.
Don’t get me wrong, the technology is nice but I don’t see myself consciously enjoying AI music, at least not till I start dementing or smth.
echelon
This is where AI becomes a tool for the artists.
We need an AI DAW, not Suno.
popalchemist
Suno is for people who would otherwise not make music. The grandchild who wants to send grandma a funny song. The marketer who wants a quick solution for a social post. There are valid reasons for both types of AI products.
megadata
A bit off topic, but is there any good recent Image to midi/XML/Sibelius stuff out there?
lioeters
Soundslice has a fairly new feature for scanning PDF or image into a music score, with export to MusicXML and MIDI.
zoogeny
I follow a YouTube channel Gamma1734 [1] where the creator who runs it plays a bunch of piano sheet music from the past, a lot of it in the Romantic style. The artists played on the channel are all obscure to me. None of the names I consider "masters" from that period or style. And to be frank, the average quality of the compositions is not great.
My own opinion, as an amateur musician with no classical training, is the music I heard on this AI generated page would not have stood out to me compared to the rest of the pieces on that channel. I would wager that in a randomized test of these average middle of the road composers vs. an AI score generated by NoteGen that almost no average person would be able to distinguish them.
That leaves two categories. The highly trained individuals (e.g. professional composers or well educated amateurs) who may be able to tell a difference (I would love a real study on this and not just "vibes"). And the master composers (Beethoven, Chopin, Brahms, Liszt, etc.) which stand above the average human composers of their own times.
I would admit the pieces composed by the AI aren't at the level of the master, but they do feel indistinguishable to me from the average.
TaupeRanger
Even the cherry picked examples are boring milquetoast? Maybe this is an example of "this is worst version of this technology" but I'm still skeptical of any generative system that isn't publicly available for testing.
The major issue I can foresee going forward is that the only public domain sheet music available is in the classical idiom. And people don't really care about writing new pieces in a Baroque style. It's a fun exercise, but not really relevant to modern music. The "pop music" listed in the article sounds like "Baroque-pop" presumably because that's what is in the training data. That's why a DAW style system would be more useful and welcome, although there again - what data do you train on? There aren't just millions of publicly available "EDM DAW files" laying around to train on.
zozbot234
> The major issue I can foresee going forward is that the only public domain sheet music available is in the classical idiom.
I'm not sure that this is accurate. You can take a look at IMSLP and see how much music there is that's in a more "popular" style (while still not anywhere close to modern pop, of course). Especially vocal music (most commonly songs) and instrumental dance pieces, which were the most common choices among 19th century amateur performers playing music in their home salons and parlors. It takes a lot of work to accurately transcribe the music from score notation into a form that AI can be trained on, but that work too is happening via e.g. the "OpenScore" effort. So I think we can expect better AI-generated notated music in the future. It could become a nice "toy" domain for sequence modeling (including language modeling) architectures, where high-quality results might be achievable with very little training effort. A bit like MINST is today for image recognition.
TaupeRanger
Is this an AI generated response? Spending 60 seconds on Google reveals every argument to be nonsense. A simple Google search for "OpenScore" reveals a MuseScore page with only 14 scores (all of which already exist in IMLSP). So what kind of "work" you think is happening there...I haven't a clue. Searching IMLSP for popular songs reveals mostly recordings and badly photo-copied PDF scores - nothing usable in the MIDI, XML, or ABC formats.
jnwatson
This is notably better than previous attempts. I hear the basic structures that make a piece of music more than a series of notes.
None of the pieces I heard were good, but I think they'd get a passing grade in a composition class.
Rochus
Here is a great online service using NotaGen where you can directly listen to the pieces and follow the scores on screen: https://piano.fm/
rerdavies
Notes in the left and right hand that occur on the same beat don't line up properly.
The human performance at the bottom of the page was nice. But you can also give a great musician a zipper and paper bag and they can make it sound nice.
It’s just like LLM generated prose to me. The orchestration and motif development is musical sounding, but there’s these artifacts that are like the “It’s important to consider” or “Would you like further suggestions” of the aural world.
I think it would be fun to write up a detailed analysis of a full piece.