Emergence of Original Music Through Algorithms
nautil.us
nautil.us
Wikipedia:
> Scott developed some of the first devices capable of producing a series of electronic tones automatically in sequence. He later credited himself as being the inventor of the polyphonic sequencer. (It should be noted that his electromechanical devices, some with motors moving photocells past lights, bore little resemblance to the all-electronic sequencers of the late sixties.) He began working on a machine he said composed using artificial intelligence. The Electronium, as Scott called it, with its vast array of knobs, buttons and patch panels is considered the first self-composing synthesizer.
https://en.wikipedia.org/wiki/Electronium
You will know his popular music from the warnerbrothers cartoons :) https://youtu.be/YfDqR4fqIWE
The repetition makes it sound a lot more musical than some of the more extreme academic music you can find. But it's still a product of its source material enhanced by human editing.
There's no sense in which this is any kind of AI. There's no NN, no symbol grammar, and barely even the simplest rule-based expert system. It's more like a semi-random razor blade tape editor and mixer.
But... it's new to most audiences, who won't have heard of Autechre's more interesting experiments with Max - or any number of other projects - and who are used to a rather precious narrative about the sublime nature of musical composition, which the text shovels on with a back hoe.
It's one of the odd things about music - it's performative in more than one way. Not only are composers supposed to perform music in the obvious sense, they're also supposed to perform "I am a composer and/or performer" in very standardised genre-specific ways.
In academic music and other genres aimed at intellectual audiences that means the music has to be supported - i.e. marketed - with program notes that suggest to the listener how important, culturally significant, original, refined, and creative the music is.
If you take away the notes and let the music speak for itself, you immediately lose a lot of the impact.
I suspect most people who heard on this Soundcloud without any explanation would think "That was weird and maybe mildly interesting" and move on without giving it a lot more thought.
AI won't have solved music until the music can stand on its own without needing that kind of marketing.
David Cope's work is interesting in this regard. His algorithms composed sheet music that is played by a pianist. He's done some blind tests where the audience appreciated the "work of Chopin" until they came to know the truth.
I had some difficulty taking the rest of the article seriously after reading this. I know music is subjective, but I think the author may have been projecting somewhat.
Don't forget, this is a three year labor of engineering love for the author!
The author's description of the music is slightly corny but I had a very similar awestruck and eerie feeling after the first time a python script I wrote generated some pretty passable electronic music.
Before you scoff, I am a trained classical guitarist and play a variety of instruments. To be clear, nobody needs my permission to enjoy it. I found nothing that I enjoyed.
I didn't find anything emotional about them. Other than the concept being interesting, I've not found anything else to appreciate.
A tech note: I would prefer if I didn't have to install the SoundCloud app to listen to the samples on my phone.