The Beatles and AI

[jazzereh]

I guess from reading the many comments above that the headline is more clickbait than anything. Sounds like this is nothing new for music production. My takeaway from the article was that AI had been used to actually generate the vocal. That is where I get into the uncharted and potentially disastrous aspects of AI. Probably the music business is one of the less to be seriously impacted by AI as it evolves.

[emanresu]

I guess from reading the many comments above that the headline is more clickbait than anything. Sounds like this is nothing new for music production. My takeaway from the article was that AI had been used to actually generate the vocal. That is where I get into the uncharted and potentially disastrous aspects of AI. Probably the music business is one of the less to be seriously impacted by AI as it evolves.

The future of music business.
Imagine a hall with 1000 people with headphones having to push one button.
Whenever the AI music feed wakes one up from the coma of boredom, the person pushes the button.
That's how we'll get our new music very soon.

[jameslovestal]

I guess from reading the many comments above that the headline is more clickbait than anything. Sounds like this is nothing new for music production. My takeaway from the article was that AI had been used to actually generate the vocal. That is where I get into the uncharted and potentially disastrous aspects of AI. Probably the music business is one of the less to be seriously impacted by AI as it evolves.

I was wondering the same thing about how much actual "AI" was being used here. E.g. is there software that can take an actual singer's recorded-vocals and then create a new song, that has different lyrics (words) then the singer has ever sung?

E.g. Lennon singing All of Me. The same thing could be done for solos. This is way beyond sampling and once this was perfected it could have a very seriously impact on the requirements for \ employment of, human musicians.

[starjasmine]

I was wondering the same thing about how much actual "AI" was being used here. E.g. is there software that can take an actual singer's recorded-vocals and then create a new song, that has different lyrics (words) then the singer has ever sung?

Apparently, the answer is "yes." In this podcast (which I posted on JGF about a week ago to virtually no response) there is a discussion of AI-created "soundalike" tracks being posted to Spotify that use AI to model a popular vocal. (I believe the example was that a Swedish company created a soundalike track using an AI model of pop singer Drake's voice.)

I haven't heard the whole thing yet, but this is really interesting:

How Musicians are Navigating Streaming Algorithms, AI and Automation | KQED

UPDATE: Some info cherry-picked from the first ten minutes or so of the podcast:
- 57,000 artists on Spotify are paid over $10k/year for their stream plays. The rest make less than that.
- This is less than 1% of all artists on Spotify.
- The label takes 50-80% of that revenue.
- If the artist is a band with four members, that 50% of 10k gets split four ways; i.e., each bandmember grosses $1250/year.
- Spotify, Amazon, Apple Music etc. do not publish their streaming rates; they are not legally bound to do so. So artists never know what they will get paid till the check comes. The aforementioned streaming providers were invited to provide a statement or actually participate in the show, but "none of them responded in time to be included."
- A number of companies located in Sweden are using AI to create copycat tracks that sound much like hit artists, but the artists see no revenue from this.
- If you want to support an artist, buy their music directly from them instead of streaming it. Use streaming to preview and then purchase the music directly from the artist.

E.g. Lennon singing All of Me. The same thing could be done for solos. This is way beyond sampling and once this was perfected it could have a very seriously impact on the requirements for \ employment of, human musicians.

There might be limitations now that advances in computing power could render moot at some point in the future. Further, if the AI vocal is trained only on one track, maybe the output of the AI-generated vocal can only deviate so much from the original track. But I would imagine that if you built the model from all of the recorded work of the human vocalist, you'd have a bigger universe of vocal artifacts from which to choose when generating a new vocal. This is an interesting (and concerning) premise when you consider who owns the out-takes from a recording session - the label? The artist? No doubt it is a convergence of existing copyright law vs what the artist's contract specifies.

Not intending to hijack the thread, but this seems related: existing copyright law specifies that when an original melody is captured on electronic media, the artist (or the person running the electronic media?) owns the copyright, even before an official copyright is filed with the US Library of Congress. So... these computer dudes generated and captured every possible sequence of 10 diatonic notes and captured that output to a hard drive. In theory, this means that they now pre-emptively own the copyright to every such melody that anyone else might compose afterward. However, he's not trying to own every melody; instead he proposes that all melodies should be free for all to use - essentially doing away with the current copyright system. This video interview with Damien Riehl is only 10 mins long; there are others that provide more detail, including his hour-long TED talk.

It's clear that the legal system is falling behind technology, especially AI, and that a new paradigm must emerge in order to protect creators, children, really everyone and anyone who may be affected by advances in technology.

OK, sorry to hijack. On the OP, I'm interested to hear the "new" Beatles song, but I think if it was that great, they'd have released it back when they were still together.

[emanresu]

Thinking about REM's song "Drive".
The impact comes from the stalling in a very very "feely" way. But the notes are familiar and toyable?
It is a massive song with a touch of uniqueness to it.. which actually is the whole point.
So it is worth to try doing it again with different notes, sounds, melody....
If asking from AI to do something similar, you'd go "make something like that 1000 times over".
The poor person who has to listen to that shit.. needs some free xanax after. At least they got paid ok, hopefully.

[DawgBone]

The future of music business.
Imagine a hall with 1000 people with headphones having to push one button.
Whenever the AI music feed wakes one up from the coma of boredom, the person pushes the button.
That's how we'll get our new music very soon.

A good solar flare will solve most of the world's problems.

People said the same thing about DJ's but there is a certain attraction to live music played by actual musicians that is never satiated by a bunch of tech bullshit. Clubs want bragging rights "we have live music X nights per week at this club". People want live music. There is a whole culture in my area with people who make a point to go see live music every weekend. For some people that is their passion in life. In fact I'd say that these people have put a lot of food on my table. They aren't interested in A.I. It's about live music, socializing with others. If you'd like to eff up the world of those around you you take away live music and the socializing that accompanies it and disallow attendance of religious services and see what kind of dark and depressing shithole realm you've now created. Oh wait....

[starjasmine]

Yeah, I get a certain kind of inspiration from seeing great musicians perform live. The inspiration I get from a well-written recorded song is entirely different. I guess you could say it's aspirational to see someone do something great live, without studio tricks. "If (s)he can do it, then so can I (try to) do it."

But most people do not have the same relationship with music that a musician does. The average consumer treats music like audio wallpaper and they don't (maybe can't) distinguish inspired playing from generated soundalike stuff. It is both the interesting harmony and high musicianship in jazz that sets it apart (for me) from most of the dreck that passes for pop these days. I still like certain pop music, but there's literally no mystery in most of it for me.

[st.bede]

There is a professor at UCSC that has been working on computer generated music for years. I heard one of the the computer program created piece of music. It was a movement in the style of Bach. If I had not been forewarned, I do not think I would have been able to tell that it was not Bach. I listened for a lack of humanness. I think I heard a little rigidity in note choices, but I am not sure. That was eight years ago.

The question of what it means to be human is difficult to answer. Often it is a sense of imperfection or limited knowledge. These limits can lead to wisdom, but often they lead to hubris.

In the future, maybe our shared level of empathy will be what defines us. However, at this moment there is too much greed, and selfishness to allow for our collective ability to thrive, or maybe even survive.

Then again it is as much human to feel fear and anger.

Ultimately the choice will be about living in fear, or living with a concern for others.

So far it is not looking good.

[supersoul]

Not intending to hijack the thread, but this seems related: existing copyright law specifies that when an original melody is captured on electronic media, the artist (or the person running the electronic media?) owns the copyright, even before an official copyright is filed with the US Library of Congress. So... these computer dudes generated and captured every possible sequence of 10 diatonic notes and captured that output to a hard drive. In theory, this means that they now pre-emptively own the copyright to every such melody that anyone else might compose afterward....

It's clear that the legal system is falling behind technology, especially AI, and that a new paradigm must emerge in order to protect creators, children, really everyone and anyone who may be affected by advances in technology.

Recent cases have come up against granting copyright protection to AI generated imagery.
AI-created images lose U.S. copyrights in test for new technology | Reuters
New US copyright rules protect only AI art with ‘human authorship’

Even the US Supreme Court weighed in, on patents for AI generated designs.
US Supreme Court rejects computer scientist'''s lawsuit over AI-generated inventions | Reuters

I think they might be going too far, because artists/musicians can use AI as a tool to generate raw material, that is then used to make something. Does it not get copyright protection because it was originates with AI. People write lyrics with random word generators, are those lyrics not protected?

Anyway, here's the Beatles singing Radiohead: Beatles sing Radiohead AI

[Stevebol]

5 minutes for free;

Vocal Remover & Instrumental AI Splitter | LALAL.AI

[Jimmy Mack]

With the quality of POP music today (or the lack thereof), does it matter? If the new JL song isn't smashing, I won't care. Will you?