Live data from Hacker News

DiffRhythm: Fast End-to-End Full-Length Song Generation with Latent Diffusion

aslp-lab.github.io

91–100 of 121 posts

Re: DiffRhythm: Fast End-to-End Full-Length Song Generation with Latent Diffusion

#91

Earlier quoted context omitted.

You don’t understand how the ‘bonkers’ cost of creative worker relates to businesses that make creative products?

No, I only understand that you're doing your best to be hostile and strawman me. So please, do kindly explain.

I’ll get right on that

Re: DiffRhythm: Fast End-to-End Full-Length Song Generation with Latent Diffusion

#92
post #8

This is primarily architecturally interesting in my opinion. Output songs have unusual noticeable artifacts, and I would guess they become more noticeable the more you listen. That said, wow. An end to end FAST architecture that can infer a 4.5 minute song in 10 seconds is a compelling thing. I didn’t see if we got open weights, but my guess is that this is not crazy challenging to train, and some v2/v3 versions of t…

The huge missing issue is direction. Songs are way more than just a 10 second style reference and lyrics. Even the most generic pop song from the 90s had recognizable choruses some repeated bars and some ebb and flow to the song that connected to the lyrics to make it interesting to the human ear. Right now the generated songs, as you noted, somewhat glitchy lyrics over a bland backing track that just sort of goes at one speed and note for the whole of the lyrics.

Re: DiffRhythm: Fast End-to-End Full-Length Song Generation with Latent Diffusion

#93

None of this is music. It is noise that sounds likes music. Pretty analogous to how AI slop is not information, but just words that are arranged to look like information.

Makes me think of the demons from Frieren who are monsters that have learned to speak and act human to hunt people better.

Re: DiffRhythm: Fast End-to-End Full-Length Song Generation with Latent Diffusion

#94
post #44
post #29

Earlier quoted context omitted.

The ability to record has led the greatest expansion I musical artistry in human history. Ty it don’t think peasants were listening t to Bach, do you? Only the extraordinarily wealthy could afford to have music as anything like an every day thing.

Lots of common people did listen to Bach, because he wrote many works for church organ. Church attendance was almost universal, and even small churches had (small) pipe organs.

His work was not commonly performed in his lifetime, and I think you're rather proving my own point? Yes, they could perhaps occasionally listen to Bach, if the organist at their church was aware of him (most would not have been, not until hundreds of years later), had the music, were willing to perform it, and you happened to be in attendance when they did. That's a lot of chained ands.

Re: DiffRhythm: Fast End-to-End Full-Length Song Generation with Latent Diffusion

#95
post #84

Earlier quoted context omitted.

The tools and knowledge for making music are already unbelievably accessible. Anyone with an internet connection and a decent computer can read about music theory, learn to use a DAW, and get some basic virtual instruments. The same goes for producing art, which doesn't even require anything digital. This does not augment the music making process in any way, it simply replaces it with what might as well be a gacha ga…

Sure, with several hundred hours to spare one can make some songs in a DAW. Now one can make something as good/bad in maybe 1/10x the time. Or, given the same time investment, one can possibly make something better!

The goal of AI automating labor should be to give us more leisure time to pursue hobbies, not to fill our limited leisure time with low quality substitutes for those hobbies.

Making an activity in which the primary limiting factors for most people are the time, knowledge, and effort required (as opposed to expensive tools) into an effortless slot machine pull is enfeebling to human creativity and agency. Who will spend the hours of making bad music to get to the point where they become good if they can just rely on something else to generate music that's "good enough"?

There's something to be said about all this which is related to AI generated images that I rarely see brought up: people with specific skills play roles within groups, so AI making their hobby that they dedicated so much time to more easily accessible makes them lose social value, which might make them quit altogether.

The common response that "people should make art because they love it, not for attention" is a prescriptive statement that supposes there are more or less "pure" forms of performing an activity and also ignores that art is a form of communication.

Re: DiffRhythm: Fast End-to-End Full-Length Song Generation with Latent Diffusion

#96
post #52

Earlier quoted context omitted.

This is why we never should have invented the phonograph. People who want to listen to music can just buy a record, making it literally impossible for them to perform an activity humans ENJOY doing. Without it everyone would surely be making all their own music, and nothing valuable would be lost

There are like 6 core activities that bind humans together: shared creation of food, myth and music; co habitation, protection, child rearing. We've done these things ourselves for hundreds of thousands of years. As we are increasingly convinced to buy them for convenience we loose the very things that make us know our connectedness. So ya, there are real problems caused by the convenience of technology

People will still enjoy making music. Musicians will make music quite regardless of whether anyone is listening or whether there’s recordings or AI available.

Re: DiffRhythm: Fast End-to-End Full-Length Song Generation with Latent Diffusion

#98

Cool. Obviously needs some work. Lots of artifacts. Something to build on though. Lots of sour grapes comments from folks. Too bad. Not what I expect out of Hacker News. Glad people are pushing the technological envelope and exploring this space despite the strong negative emotions.

[flagged]

Can you please make your substantive points without snark or fulmination? We're really trying for something else here.

This is in the site guidelines: https://news.ycombinator.com/newsguidelines.html.

Re: DiffRhythm: Fast End-to-End Full-Length Song Generation with Latent Diffusion

#99

Earlier quoted context omitted.

Whats the end game here? Let’s follow the AI and automation craze to its eventual conclusion - automations everywhere, humans are either employed in automation industry, or are unemployed at a massive scale. Stable jobs are replaced by ever-optimized gig economy for some, and chronic poverty for others. For there to even be economy - the massive underemployed population subsists on government welfare. Cynic in me thi…

> Cynic in me thinks that all of the wealth generated by enormous productivity gains resulting from automation will not find its way towards population displaced by it. Empirically, that's not true. Unemployment was at an all-time low after most of those jobs were eliminated, and wages after adjusting for inflation continued to rise in real terms.

I am inclined to doubt the sources of these empirical observations. Statistics are funny like that, “average patient temperature in the hospital” effect and frequent inability to correctly attribute confounding factors outside of observed window.

Equally bad is anecdotal evidence, but I’ll drop some anyway. For a while now I am observing a crisis thats, admittedly subjectively, easy to see - but is somehow absent in those empirical sources citing economic accomplishments. An indirect evidence of what I am talking about - is crushing defeat of democrats/establishment in last election, following among other reasons, quite a backlash for boasting about said accomplishments.

But rather than picking issue with one of my points - I still would like someone to describe the counterpoint to my dystopian expectations - where, for example, would all those professional drivers I mentioned earlier go?

Ps. Oh speaking of statistics - remember Greenspan’s “there’s no real estate bubble, there’s froth in individual markets” right before 2008 financial crisis? It be funny like that, sometimes much derided common sense is all you need /shrug.

Re: DiffRhythm: Fast End-to-End Full-Length Song Generation with Latent Diffusion

#100

Earlier quoted context omitted.

Business doesn’t hate creatives, and is not specifically targeting creatives to automate them away. Any job that can be done as good for a lower price or better for the same price is going to be a target.

> Any job that can be done as good for a lower price or better for the same price is going to be a target. So they just hate humanity in general then.

They don't hate at all. They are just maximising profit (which they have an obligation to do). If they didn't replace you with more efficient things, they would be outcompeted and die.

So, feel free to criticise capitalism and how inhumane it is, but don't anthropomorphise it by ascribing human emotions to the system.

Post reply on HN