The embattled National Public Radio in the United States happily (in spite of draconian funding cuts) still pursues intelligent discussions on a huge variety of relevant issues present, past, and evolving. Such is the case with a recent series of discussions (some shorter, some longer) that have focused on AI generated music. Of course the primary genre in these discussions is that of “pop” (read “economically viable”). But what of all the music that makes no money? Like mine, lol, but they make me happy.
My point here is not to denigrate NPR, rather to append a perspective on my musings about new classical music and AI. Since I am hardly an expert on this, all I can reasonably do is provide an interesting example that I have stumbled upon.
OK, this is probably my first encounter with AI in music, at least as far as I am aware. It is an album on Bandcamp by a friend of mine. Henry Lowengard is a musician, actor, instrument builder, programmer, hacker, and performer. He was interviewed by another friend of mine who has a radio program out of Kingston, New York. Here is the fascinating program that aired last year in which program host Peter Wetzler interviewed Mr. Lowengard about this album.
Peter is a composer, pianist, and a driving force in the arts community that is Kingston. His weekly show features music from all times and all places. He frequently plays new music and has introduced this writer to a great deal of music.
So, to the album at hand. Give it a listen. It’s a fascinating album that puts this listener in the mind of composers like Ken Nordine, Robert Ashley, Pamela Z, Charles Amirkhanian, and Laurie Anderson. Not that other listeners will have the same experience.
Give a listen to that broadcast in which Henry talks about how he made the album. That helped me to grasp what was going on. It reminded me of the free floating improvisational texts that comprise Ashley’s operas, in particular, Perfect Lives. The pieces on Lowengard’s album are stream of consciousness. But whose consciousness? Henry’s or AI’s?
Apparently, Henry uses various programs and programs via the large language model of AI. His commands instruct it to produce some kind of music with some kind of texts. He is not actually writing the music or the texts. Rather he is asking AI to interpret his very linguistic ideas to produce both text and sound.
This album surely has its antecedents but, in many ways it also sounds like nothing I’ve heard before. If you like creative experimentation, give this a listen. It is but one incarnation of AI non-pop music. I doubt it will crack the Billboard charts but it made it into my listening queue and I’m glad it did.
New Music Buff 
Hi there Allan! Thanks for the highlight!
Large Language’s text’s are all generated with GPT-2, which I run on my laptop. GPT-2 is more dream like and free associative than more “useful” AI agents. I give it a shot prompt which it more of less ignores, and get it to generate 50 responses or so. I then go through them and see if any of them either make sense or have some literary property worth exploiting. Sometimes, there seems to be a need for related musical material, so I use tools like Suno and Udio to musicalize some of that text. Usually, for the background beds of the pieces, I generate music using musical tools like Senode, or I’m playing live (like the squeaky table or the bicycle spokes).
Large Language also has an album of things produced when Suno decided to ignore what I told it and made up songs in a language that nobody speaks.
[ https://largelanguage.bandcamp.com/album/largely-incwvugble ]
Mostly they are left alone, but I added some tracks (recorders, Omnichord) to “The Momus of Mnezhige Thistle” to make it sound even more like some lost Scandinavian kid’s TV show. I got ti to make some end title music for it too. At some point the tools will be good enough to make a whole 1/2 hour show.
Suno is doing some great work orchestrating and playing music and voices. Here, I gave it a short phrase or music I made on my Amiga in 1986 (!) and told it to come up with different styles, with singers just going “pa,pa,pa, da,da,da” for lyrics. It turns out, that’s pretty fruitful! There are 64 variations in this playlist, I just had to stop! [
https://jhhl.net/r/spl-as ]
I’m hoping to move more and more of this processing in-house so I can train it on my musical sources and noises and interact with it more!
LikeLike
You’re most welcome. I do appreciate the “liner notes”. I’ve listened to the Radio Kingston episode multiple times. That clarified a lot for me.
One of the problems I have when trying to grasp a new piece is that some music is nice and linear. But now, music has wider definitions. Material like your album, much of Robert Ashley’s operas, and others disturb my analytical brain and my focus wanders. Now, pretty much any music can transport the listener, but this stuff can be very mind manifesting. I think that the frequently obtuse texts trigger ways of thinking. Some of the ideas expressed along with the disruption of learned grammatical rules, force attention to the “errors” (whether perceived or actual) and this takes away my attention a bit. But the process is ultimately mind expanding in some ways. And it’s worth the effort to at least listen more than once. To be honest, I’m still mostly lost, but enjoying the journey through your album. That’s pretty much been the case for me with Ashley, PamelaZ, etc., all artists I greatly admire. Some family issues have kept me more off the grid but I hope to be back up to speed on my writings. Talk to you soon. Peace, health, and music to all, and to all an unsilent night.
LikeLike
You’ll probably like to hear the Joe Frank material I reference (Peter didn’t play a particularly good example because he knows nothing about him).
I actually put together the first Joe Frank Webpage, not as useful as Joe’s own website. https://joefrank.com
There are mostly excerpts there, but it gives you a better idea.
I wish his heirs would release some more full length shows, but there are some radio stations still running them (!) .
LikeLike