AI music

what a clickbait letdown. AI ""music"".

it'd be wild if AI really could write music someday though

Not that hard for AI to make a compelling replica when so much music on the charts is trivial bullshit. Half of it can't even be bothered to use four chords, it's just a single bar on loop with three notes of vocals over it. No chorus, bridge, or anything. This actual Drake song is indistinguishable from the AI track, and not in a good way.

This is not really AI music, but some Oasis band members wrote music and just had AI mimic Liam's singing. It is kind of crazy how good it sounds.


I am not sure how I feel about it. On one hand I think it would be awesome if say AiC wrote some new music and used it for vocals. Would that just feel disrespectful though? Dunno, but I could totally see that sort of thing happening soon if not already. Or even plugging your favorate singer in over other bands songs.

Captain Suave said:

Not that hard for AI to make a compelling replica when so much music on the charts is trivial bullshit. Half of it can't even be bothered to use four chords, it's just a single bar on loop with three notes of vocals over it. No chorus, bridge, or anything. This actual Drake song is indistinguishable from the AI track, and not in a good way.

Click to expand...

Thats my issue with it.

In the "music" in question, the "artist" is neither playing an instrument nor singing. Its literally not music, by any reasonably definition, any more than Dylan Mulvaney is a woman. But we're asked to be impressed that AI has matched this "music".

I'll be legitimately impressed when AI makes music, which is an art form involving singing and musical instruments. I'm not even skeptical it can happen. Just pointing out that in no way does this qualify.

Not really music, but I think this channel uses an AI voice and it is pretty good


It sounds better than those celebrity voice AIs that were running around on twitch awhile back.

Tuco said:

music​

noun​

mu·sic ˈmyü-zik

1a: vocal, instrumental, or mechanical sounds having rhythm, melody, or harmony
choral music
piano music
recorded music

b: the science or art of ordering tones or sounds in succession, in combination, and in temporal relationships to produce a composition having unity and continuity


I'm not sure I understand your argument and I don't really care one way or the other, but AI-generated music that involves no physical medium other than the computer it's run on is as qualified as music as my 4 year old banging a rock on my living room painted drywall to the beat of "baby shark" or Mozart.

Click to expand...

I cant believe you're resorting to a dictionary to argue that Drake counts as music. We all get that one some kind of formal bullshit definition basis you can play debate club and argue that what Drake does technically count as music.

That doesnt change the reality that every one of us experiences in the first few second of hearing it.

Nor does it address the bigger issue, that its just not impressive at all when AI mimics bullshit "technically music but not really" like this. When AI makes music that human beings hear, and think "thats music", that will be sincerely impressive. This was a cheap trick, a technicality that doesnt really count on any meaningful level.

I've been messing around with this but not had much luck yet:


Useful forks had one with a jupyter notebook and I got as far as the training and then there was some sort of dimensionality problem I can't fathom.

Being able to generate music would be badass for game making.

I think this is the original paper with lots of samples: MusicLM

This fork is archived, but the training phase is working, though It has been running almost 2 hours.


I had to kill everything and turn off the compositor to get enough vram to train.

Took about 5 hours to reach the halfway point, assuming I'm reading the python correctly. Seems to want half a million training steps. I'm on 288k.

I guess I should just be happy this is working at all on my crappy old 980.

Training finished! Aaaaaaaaaaaaaaaaaaaaaand I have no idea where it went. Nor any clue what to do with it if I had it. I kind of think it was just in memory and when the program finished it just vanished into the ether.

Or maybe it is off in some strange directory I'll never find it, forever consuming a few gigs of space. That's what I get for attempting some random archived fork.

Gave this one a go: GitHub - zhvng/open-musiclm: Implementation of MusicLM, a text to music model published by Google Research, with a few modifications.

It has some pretrained models, but they don't work and hasn't really been any response in the issues section.


There's a large and small model and the large doesn't fit in my 4 gig card. I'm guessing that's what the problem is. No idea how to adjust it to fit the small, but it might work for one of you with a gpu from this century.

All my AI fiddling with here and other audio and image stuff has really revealed to me what a steaming pile the entire python ecosystem is. Not just the language but the open sores garbage libraries just piled in heaps atop each other like a shitty chinese apartment block.

The versioning is such a nightmare. Blortus needs >4.6 and <4.23 but blortus also needs blort which needs < 4.3! Endless detangling of noodly garbage. What is going on with the language itself when stuff is so sensitive to versions? Is it just api changes wildly out of control? Is the repo just open to any idiot to commit?