100%. However, coming from a baroque background, I’m perfectly happy composing the main line, continuo, and letting the players take it from there.
In fact, I was recently listening to a phenomenal performance of Handel’s Israel in Egypt, by Harry Christophers and the Sixteen. The “They Smote” chorus struck me in particular. Extremely powerful, but quite a far cry from the relatively simple score, with all those drums! Almost a “cover” version, TBH, but perfectly legitimate. Handel would be happy, I imagine! Here’s the link: Israel in Egypt, HWV 54 : He smote all the first-born of Egypt
Not sure I’m understanding your frown. What I meant is that baroque composers would write the barebones as I describe, and hand that off. Of course, in this case, the players are AI…
I don’t understand, this a very good performance by a choir of fantastic singers and an orchestra, specialised in this kind of music (and using instruments close to what Handel would have found at his time).
And of course a good conductor, not to forget..
It’s all original Handel (Händel), the instrumentation with trombones, all original. There are no drums in this chorus - but isn’t it amazing, how a good performance can trigger one’s imagination.
As Janus points out, this is the first publication (by Randall in London ca. 1770).
That’s a long time after Handels first performance of this oratorio (in 1739).
You can read up some very interesting fact on this page:
Here is an 18th-century copy of the original trombone parts for “Israel in Egypt” He smote (trombones).pdf (844.3 KB)
There is also this manuscript from 1760, which also doesn’t include trombones in the score; however, trombones were included in orchestral parts. This was consistent with the common practice, as a fully orchestrated conductor’s score didn’t exist at that time.
(The trombone part you uploaded comes from this very source)
Yes, it is. Very clever at absorbing the work of others and applying all those bits and pieces is (mostly) musically appropriate ways. It cannot come up with anything new, but neither can 99% of the working musicians out there.
I was also curious about the capabilities of Suno. I did not feed it anything but a simple text description. I believe what I asked was something like “Produce a song in an island funk, Afro-Cuban style with trombones, on the subject of an unsuccessful fishing trip.”
Within 3 minutes it did have several songs, one of which was kinda catchy. What it produced was more like a New Orleans style. The lyrics were a bit goofy, the horns were pretty boring ostinato style and some of it used lines that were in a mode that was out of context. But the overall vibe was good party music. I have rearranged this for a big band and we will play it on our next show, along with some commentary from the stage about the state of AI. While the OP’s rendering was pretty close to a finished work, I’d like to think our band arrangement of my Suno piece is considerably better and more interesting than what Suno produced.
Nonetheless, it is scary just how good it can be. My attitude is that AI is an unstoppable force, mainly because the tech bros, as always, are years ahead of the legislators. They were able already to steal a massive amount of copyrighted material with no consequences. I don’t think you can put that genie back in the bottle.
In my case, I can’t say that Suno is ripping off any particular artist. The song it produced sounds like every horn street band in New Orleans, and I guess they have all been copying each other for years. AI can steal from musicians faster than musicians can steal from each other.
I wonder if Suno uses an intermediate layer of MIDI, or if it just pastes together sound fragments, more or less like Band-in-a-box does. If Suno could provide this orchestration in MIDI or MusicXML, that would be worth something. I doubt that I would keep the majority of an orchestration, but if it could save me the first 50% of the orchestration/arranging job, I’d pay good money for that.
For me that alone is a deal breaker. I work so hard at composing and arranging with precision that unless I’m writing something aleatoric or with instructions to improvise, I wouldn’t want my work changed that much. And if I were writing something with aleatoric passages, I have my doubts Suno could handle tasteful improvisation well.
It’s a nice little composition by the way, but it certainly is period-style idiomatic. AI models are well trained on historical music, so I think they handle this well in general. I do wonder how frustrating it would get for anyone that writes in modern style, extended techniques, etc - styles where the structure and harmony is less statistically predictable and able to be captured by an algorithm.
I assume you mean for creating mockups based on this style of music? Obviously DAWs will have a place in the future for performing musicians, pop artists, music producers. The best AI tools will be ones that support them in this work, not simply cover their music.
All that said, I do appreciate you sharing this - good food for thought. I’m working on a commercial project right now with these string legatos and I own a lot of the major player VSTi string libraries and none of them are sounding good to me (in Cubase). I wish I had the budget to hire a live string orchestra haha. The string arrangement is mostly background to a larger pop production, so I think I’m going to at least try to see if I can get Suno to take my current strings and spit out something more realistic and lush. Because the strings are background, in spite of everything I said above I would be sort of ok if it changes my arrangement slightly - as long as it stays in key and has decent voice leading. I really hate how synthetic string libraries can sound with legato sometimes, so perhaps that would be the lesser of two evils in this context…
I hate AI music as much as anyone here. My non-musical friends are always sending me AI music and telling me how awesome it is, and it kills me. But it doesn’t matter what I think. AI is here.
But I for one am not all gloom-and-doom about it. You can’t force the masses to listen to the good stuff. Creators need to just keep creating and hope for the best. Good art will prevail in the end.
And I just want to clarify my point: My point is that I thought until recently that “AI music” means you “describe” the sort of music you want it to make, and it composes the entire thing for you.
That is not true, and I thought it would interest people here (although I may well be the last to have realized this). You can, in fact, feed AI with a demo version of your music, and, depending on more factors than I understand or can enumerate, AI may well produce a version of the music that is perhaps 85% faithful to the original, and the reproduction quality of which far surpasses anything an average person can achieve even with a DAW, let alone Dorico alone.
Given the rate of progress of these things, I would not be surprised if in a year or two there will be an equivalent of Suno today that allows uploading actual scores and getting back a Suno-level rendering, note for note.
I’m all for live musicians, and orchestras, but this would really be the holy grail for many impecunious musicians, and I was amazed, having played around with Suno, how close to that we seem to be.
Here’s another example for anyone who’s interested (but not for anyone who isn’t).
A Theme and Variation I recently composed. The notes. And the rendition.
Again, IMO this rendition is phenomenal. I think there is no VST or DAW that can attain this level of realism. Correct me if I’m wrong.
And yes, once again there is excessive latitude being taken with the score which I wish could be eliminated. But I’m willing to compromise in exchange for the quality of the playback.
Interesting to see how that works out. I wonder if you’ll even be able to convince Suno to create a part that consist of strings alone. It always wants to add extra instruments. (Although there is also an option to export stems.)
Maybe I’m a skeptic or my standards are too high, but this does not sound realistic (as in “like real performers”) to me at all. It sounds more like a film/tv/video game soundtrack speaking purely in terms of performance quality (i.e., generated using samples, heavily EQ’d and rebalanced for stereo/surround mixdown and recording quality). This is not intrinsically a bad thing; I don’t think AI-generated music should sound like acoustic music, but rather it has the most potential if it takes advantage of the electronic medium and the machine learning environment in some way.
For example if you’re creating baroque style music like this a fairly basic use case would be playing a figured bass or a theme and having the AI create variations on it alongside you, adapting in real time to any changes or new ideas you introduce. Leaving out real-time processing, machine learning could also be used to generate instruments, styles, etc. that no longer exist or are controversial (e.g., ask it to recreate a castrato for a vocal aria or Bach’s well temperament for a harpsichord piece) or hypothetical ones that could only have existed in alternate histories etc. Music in general, like other creative fields, is a place where AI hallucinations and “inaccuracies” are arguably the most interesting thing they can generate.
However, trying to have fine-grained control over anything an AI generates seems like an inevitably losing battle, since AIs are designed to prevent you from having fine-grained control and therefore at present don’t have a way to process it. They take data sets with tens to hundreds of billions of parameters and run them through artificial neural networks with a set number of layers and transformers that have been trained for a set amount of cycles; they don’t necessarily retain data they’ve already generated, and as far as I know there is no point in the process where there is a partially-complete output in a form you could understand/edit. All you can do is iterate it again with different prompts and then I suppose stitch together whichever sections of the outputs you want in Audacity or whatever.
