AI Generated Music?

[Not My Name]

For a number of years now I have been hearing songs that sound like they are made up of obfuscated parts from other songs. It should be pretty obvious to anyone when parts from songs you are familiar with are used in creating these songs. If I wasn't familiar with a lot of the music used in these generated songs, I might not be able to tell it wasn't music cobbled together with parts from other songs. For example.

It seems to me that a number of songs or song parts are selected and then mashed together by AI. What kind of software is used in generating music this way? I've heard "mashups" of songs with vocals and lyrics are taken from one song and manipulated to replace the original vocals of another song. I'm assuming that software used for that is probably related to this generated music.

I've been getting more curious about this lately. Anyone here work with software like this?

I don't much like the idea of music made in this way but I see that something like this could be a useful tool in helping to come up with song arrangements for my own music. Mainly for creating drum tracks. I'd like to be able to take a couple of drum patterns and then have the software extrapolate those parts into a full track.

 

[VomitHatSteve]

So you're talking about using AI to fill in instrumentation?
That would be Band in a Box (or a sample pack or a MIDI collection - Toontrack is having their summer sale right now; lots of drum options there)

Or are you talking about extracting stems from a final mix when you can't get a hold of the originals?
That would be Splitter or Moses.

Or are you talking about using AI to do songwriting and and arrangement?
That doesn't exist.

Or are you just noticing more and more repeated patterns in popular music that suggests they must be using some kind of algorithm to derive new songs?
That's called getting old.

 

[Not My Name]

So you're talking about using AI to fill in instrumentation?
That would be Band in a Box (or a sample pack or a MIDI collection - Toontrack is having their summer sale right now; lots of drum options there)

Or are you talking about extracting stems from a final mix when you can't get a hold of the originals?
That would be Splitter or Moses.

Or are you talking about using AI to do songwriting and and arrangement?
That doesn't exist.

Or are you just noticing more and more repeated patterns in popular music that suggests they must be using some kind of algorithm to derive new songs?
That's called getting old.

I did a little more looking around. What I am talking about is apparently referred to as "deepfake" music. That is what I am curious about.

 

[rob aylestone]

Every advancement in music has had potential for drift, but the differences between AI music - of any genre and new music is based on how people like it. People like or dislike music products by some strange kind of mental process that isn't clear. That Frank Sinatra one is a good example. It's very clever, but it's also rubbish. Not technical quality, or the AI doing it's stuff - but the result is just not Frank Sinatra. The other clips show the legal hot water - 'passing off' has legal status in many countries, so while you can graft somebodies face onto another person, and re-create a famous persons voice in quite a realistic manner - the only success marker is if people like it. If they do, and money starts to change hands, then the real owners of the face or voice will start making lawyers richer.

We had this as soon as tape became editable - things created that were not real. Since then, editing has become an art in itself. The control comes from the legal framework. rights are fiercely protected because that is where money comes from - the management of intellectual rights. We've had impersonators for years - how many people have not tried a Bart Simpson or Darth Vader voice on their sat nav? If the sat nav was supplied with a voice claiming to be Bart Simpson, but wasn't - the real Bart, or his management would be in court.

If you create music and put it on YouTube, Spotify, Apple, iTunes and the others one important question asked is are there any samples used? Say yes and you might get the music rejected unless you can prove you have rights to use these people's material. It's been like this since sampling started and this AI thing is just the latest incarnation.

 

[VomitHatSteve]

I did a little more looking around. What I am talking about is apparently referred to as "deepfake" music. That is what I am curious about.

Ohhhh. You mean like these things?


That'd be OpenAI Jukebox then. I guess if I wanted to do that I'd join their discord.

Of course, if you're just talking about reverse-engineering someone else's song and making something that sounds similar (but perhaps not in a legally actionable way), that's just called "learning to produce". People have been copying each others' styles since recorded music started. There's no magic bullet for it, but we can give specific tips on which specific effects you may want to achieve. I don't know of any AI tools that will specifically help towards that end.