Velvet Sundown: Music expert struggles to work out if popular Spotify act are AI or not

[andysays]

I can't see how anyone could successfully litigate an "In the style of our client's music" case.

I vaguely remember a case some time ago when an advertiser was refused permission to use a piece of Tom Waits' music in an advert, so they got someone else to compose and perform a pastiche and used that instead.

Waits or his record company sued, though I can't remember what the result was.

ETA looks like he won

 

[RevolutionSound]

Doubt it
Mp3s sample and compress audio.
Vinyl preserves the entirety of the sound wave.

The point made in the OP video is that the AI track has been fed by compressed sound from multiple mp3 files. It can't split the tracks into component parts because it didn't "learn" from separate component parts. It would be interesting if AI was "fed" from a vinyl catalogue instead of mp3.

Vinyl is one of the worst media for audio. Terrible frequency response and dynamic range. Digital is far superior. Most people can't distinguish between MP3 at 320 and uncompressed WAV files.

 

[Aladdin]

Vinyl is one of the worst media for audio. Terrible frequency response and dynamic range. Digital is far superior. Most people can't distinguish between MP3 at 320 and uncompressed WAV files.

Did you watch the video in the OP?
My post is about that...and how they discuss how AI that has been fed with multiple files ..all mp3. So it isn't able to fully discriminate all sounds or even recognise all separate sounds. The problem being that everything fed in mp3 format is already compressed.
That's the point.
It doesn't matter what you or I can hear.
Vinyl provides for a broader and wider ranges of frequencies.
If AI was fed only from perfect vinyl recordings the subsequent music created would be better in terms of splitting sounds and all sounds would be clearer when split into component parts.

 

[RevolutionSound]

Did you watch the video in the OP?
My post is about that...and how they discuss how AI that has been fed with multiple files ..all mp3. So it isn't able to fully discriminate all sounds or even recognise all separate sounds. The problem being that everything fed in mp3 format is already compressed.
That's the point.
It doesn't matter what you or I can hear.
Vinyl provides for a broader and wider ranges of frequencies.
If AI was fed only from perfect vinyl recordings the subsequent music created would be better in terms of splitting sounds and all sounds would be clearer when split into component parts.

I understand your point, I just don't agree with it. I'm pointing out that vinyl is actually not a very good medium for reproducing audio. Digital is far superior and MP3 encoded at 320 is probably just as good as vinyl and maybe better.

 

[Aladdin]

I understand your point, I just don't agree with it. I'm pointing out that vinyl is actually not a very good medium for reproducing audio. Digital is far superior and MP3 encoded at 320 is probably just as good as vinyl and maybe better.

Aaaaargh.
I'm not talking about "audio".
This is about sound frequencies. And how AI thats only fed MP3 files does not ",pick up" breadth of sounds and can mix sounds/voices/instrumekts together instead of keeping separate "voices"/instruments" so that when it then creates a piece the sounds when separated are contaminated.

Not getting into a "mp3 better than vinyl" thing... you're entitled to your opinion.
Personal preference is that Vinyl has a breadth of sound unparalleled.

 

[RevolutionSound]

Aaaaargh.
I'm not talking about "audio".
This is about sound frequencies. And how AI thats only fed MP3 files does not ",pick up" breadth of sounds and can mix sounds/voices/instrumekts together instead of keeping separate "voices"/instruments" so that when it then creates a piece the sounds when separated are contaminated.

Not getting into a "mp3 better than vinyl" thing... you're entitled to your opinion.
Personal preference is that Vinyl has a breadth of sound unparalleled.

You are talking about audio and you're objectively wrong.

 

[elbows]

Theres no need to focus on that angle anyway, we can just listen to what results generative music AI systems are already able to achieve. There are plenty of other factors the services have had to grapple with when training and improving their AI models, which affect the results, but improvements have still happened at quite a pace in recent years.

 

[freshapple]

Doubt it
Mp3s sample and compress audio.
Vinyl preserves the entirety of the sound wave.

The point made in the OP video is that the AI track has been fed by compressed sound from multiple mp3 files. It can't split the tracks into component parts because it didn't "learn" from separate component parts. It would be interesting if AI was "fed" from a vinyl catalogue instead of mp3.

It looks like it might be possible?

How to Separate Music Tracks Online for Free

Learn how to separate audio tracks online quickly and easily! Create karaoke tracks, covers, and more. Check out Moises' blog!

moises.ai

 

[elbows]

The generative AI services also tend to offer stem separation stuff as a tool for end users, so they clearly have those systems available in-house.

They arent big on revealing how they trained their models, so I dont know what quality of source material they actually used. Its certainly not hard to get hands on lossless digital formats that are equivalent to CD (or better than CD for certain albums), but whether they actually went down that route when doing the training I cannot say.

 

[Aladdin]

You are talking about audio and you're objectively wrong.

And...
You are missing what the OP video was about.
MP3 are compressed. Right? They lose information whether you hear it or not is irrelevant. Its lost.

So which of the following will give the best AI creative outcome from the information input ?
1. AI fed compressed information
2. AI fed uncompressed information.

?

 

[Aladdin]

It looks like it might be possible?

How to Separate Music Tracks Online for Free

Learn how to separate audio tracks online quickly and easily! Create karaoke tracks, covers, and more. Check out Moises' blog!

moises.ai

Of course it's possible.
The guy in the video does it to the AI track..

And shows how shite the sounds can end up... contaminated

 

[RevolutionSound]

And...
You are missing what the OP video was about.
MP3 are compressed. Right? They lose information whether you hear it or not is irrelevant. Its lost.

So which of the following will give the best AI creative outcome from the information input ?
1. AI fed compressed information
2. AI fed uncompressed information.

?

MP3 is compressed, which means it doesn't have all audio information. However MP3 encoded at 320, it doesn't matter. Vinyl also does not record all audio information. That's the point. You seem to think that vinyl is good at reproducing audio faithfully. It isnt.

 

[Aladdin]

MP3 is compressed, which means it doesn't have all audio information. However MP3 encoded at 320, it doesn't matter. Vinyl also does not record all audio information. That's the point. You seem to think that vinyl is good at reproducing audio faithfully. It isnt.

Honestly I am not that pushed.
My personal preference is for vinyl..as a musician.

The discussion is about how AI creates music. And right now its not drawing from Vinyl recordings.

 

[RevolutionSound]

Honestly I am not that pushed.
My personal preference is for vinyl..as a musician.

The discussion is about how AI creates music. And right now its not drawing from Vinyl recordings.

It's irrelevant, that's the point.

 

[Saul Goodman]

I vaguely remember a case some time ago when an advertiser was refused permission to use a piece of Tom Waits' music in an advert, so they got someone else to compose and perform a pastiche and used that instead.

Waits or his record company sued, though I can't remember what the result was.

ETA looks like he won

That was somebody imitating his voice. The lawyers seemed to agree that imitating the music style wouldn't be a problem.

 

[xenon]

Aaaaargh.
I'm not talking about "audio".
This is about sound frequencies. And how AI thats only fed MP3 files does not ",pick up" breadth of sounds and can mix sounds/voices/instrumekts together instead of keeping separate "voices"/instruments" so that when it then creates a piece the sounds when separated are contaminated.

Not getting into a "mp3 better than vinyl" thing... you're entitled to your opinion.
Personal preference is that Vinyl has a breadth of sound unparalleled.

Haven't seen the video yet so maybe I've misunderstood. But differentiating between MP3s and vinyl in terms of quality, seperateness of instruments, to train AI on, makes no sense. For a start you'd have to digitise vinyl and there's very little audible difference between 320kpbs MP3 and wuncompressed wav. This has no baring on being able to identify different frequencies.

There are also already AI powered tools for separating mixed tracks into their constituet parts. e.g. drums, bass, vocals, guitars. Results vary. The tools will improve.

I've been tempted to try some of these to take apart songs to learn them more easily or do some experimental karaok stuff.

 

[SpookyFrank]

You're all wrong and CD is the best medium.

 

[Boris Sprinkler]

Kula Shaker were AI. These are definitely AI. Watch out it’s coming in your ears.

 

[Brownlow]

This illustrates a question I was already thinking of before I read your post.

Does the fact that music has been created by AI make any difference to those consuming it?

I was inclined to say no, and your post seems to support that.

It probably doesn't ..now.

But imagine if AI were "trained" exclusively with pre-20th century music. It'd churn out ever more samey slop. We wouldn't have the intuitive fusions of blues, celtic ballad, jazz, electronica. Cole Porter at best, but it'd never give us Miles Davis, Jimi Hendrix, Weather Report, Leftfield ..Ruddy Yurts.

Now move the thought experiment to now and a century from now..