AI Country

 

With AI music anybody can put out a country record these days. This is my first one, which is kind of odd for me, someone who is mostly an arty postmodernist guy releasing mostly ambient music. Actually, even though it sounds generic, it is postmodern in the sense that it's based on a concept, which is important now because if you use these generators the probability that songs are going to sound the same is pretty high. There are probably songs that already exist that sound exactly like these. So I get to focus on the concept, the lyric writing and working on the visuals. That's what differentiates it from other generated songs.

The lyric is a condensation of something I had written back in January about this idea of things constantly changing. The river that you went down to yesterday is a different river today, you’re a day older, that piece of steel is more rusted, and so on.. So I thought it would be kind of interesting to make an album of songs where every song has the same lyric, but it’s not the same “river”. It would be like holding an audition of 10 artists and asking them to interpret the lyrics or imagine that every time you went down to the river in that same spot, a different variation of a song would play in your headphones.

For the visuals I wanted to use images that obviously have rivers in them, and I started to look at photographs, none of which worked. So I thought I'd use the Hudson River School paintings from the 19th century, There was one painting that worked for all the songs. Canoeing on the River (artist unknown). The other painting I used was a Thomas Doughty for the final track–the one that sounds like Peter Gabriel's Mercy Street in some ways.

Some thoughts on the individual tracks:

First Time:

One of my biggest gripes about AI music are the chord changes. Many of them use the tired “All Along the Watchtower” changes (i-VII-VI – Em /D/ C). Since the productions are so slick and the playing so good you tend to not notice the lack of imagination in the harmony. If you were to sit down with an acoustic guitar and try to reproduce some of these songs they would probably sound pretty rough and they wouldn't be very interesting. I have attempted on a number of occasions to try to reproduce them and it is very difficult–as difficult as it would be for any band that wants to reproduce the sound of the record live. My objective as a writer has always been to get a song to the point where you could make a lead sheet of it with just chords and melody. It's almost impossible to do that with any of these songs, because it would change them drastically, but would refine them as well. None of them get close to the Great American Songbook kind of writing, a time when songwriting always involved music notation.

Who are all the singers and guitar players on these songs? Obviously these are not synthesized sounds. As I've said before, a song is now all about how it sounds. Sound is a song. When we listen to it as a production we tend to not notice little flaws in the way that the vocalists are phrasing, which has to do with forcing square pegs into round holes: Imagine each song section is a box and the lyric lines are pieces of foam in fixed sizes. If the pieces are too large, it will squeeze them to fit in the boxes. If the pieces are too small it will take pieces from somewhere else, regardless of whether it is verse or chorus material (7:40). The boxes are the riffs, and the lyrics are the pieces of foam.

Words are frequently garbled, e.g. “A shadows in the night”. Actually, vocalists do this as well, and you might want to leave those little things (6:47) in to give it a human quality, and perhaps maybe they build that kind of thing into these LLMs for that purpose. There have been other songs I generated where entire verses were just nonsense scat and it worked nicely.

Second Time

Vocals are generally far too rubato and loose, disconnected from the groove. But in terms of reproducing them live, that kind of thing might make the music more spontaneous and improvised, rather than just reproduced off of sheet music or as it was on the record.

There are sometimes very beautiful moments that always capture my intention, the way that a guitar riff happens seemingly spontaneously. (12:50)

Sixth Time

This is essentially a minor blues done in a '70s soul style a la Bill Withers. It would also be a good song for Sting to do because Withers was one of his influences. This is one of the tracks where there is a lost harmonic opportunity. If there was a jazz keyboardist on this, I would ask them to use chord extensions to make it sound more dissonant and edgy. This is not something that AI music is good at at the moment, and I wonder how they're going to do that. Generating music is only a matter of choosing options, which is different from playing a chord on a piano. These AI tracks are all illusions, but most recorded music is an illusion. It used an augmented chord once in this song (15:43), but wasn’t repeated. The harmonic sophistication isn't there because everything is being generated from samples. These aren't humans playing so it wouldn't “think” of doing this kind of thing. 

Seventh Time

Actually a decent melody, albeit auto-tuned, and all chord tones with two leaps in it. Most singers probably wouldn't sing it this way.