Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Open-sourcing AudioCraft: Generative AI for audio

ai.meta.com

Post reply on HN
Print view View original

271–280 of 335 posts

Re: Open-sourcing AudioCraft: Generative AI for audio

#271

Post by munificent »

oceanplexian wrote: post #245
munificent wrote: post #188

Earlier quoted context omitted.

> the goal of our human pursuit is to advance our technologies and systems to the point we can all live carefree lives where the focus of our pursuits is self defined. Agreed! > the future should not be about preserving some peoples rights to limit our collective advancements over their personal wants. Systems that don't allow people to extract value from the hard work they put into collective advancement do not seem…

> No one's going to spend all day making candy and put it in the "free candy" bowl when that one asshole kid down the street just takes all of the candy out of the bowl every single day. Software is an infinite candy bowl. Taking candy out of the bowl does not take any candy from the person who made it. Imagine if this was the physical world, and you had a machine that could end world hunger. You could copy food like…

> Software is an infinite candy bowl. Taking candy out of the bowl does not take any candy from the person who made it.

Software is not a finite candy bowl. It is also not an infinite candy bowl. It's not like physical goods at all, not even like physical goods that can be magically cloned. It's just different, entirely.

The incentive and value structures around data creation and use just can't be directly mapped to physical goods. You have to look at them as they actually are and understand them directly, not by way of analogies.

Why do people make software and give it out for free? Is it purely from the joy of creation? Sure, that's part of it. The desire to make the world better? Probably some of that too. Are those forces enough to explain all open source contribution?

Definitely not. Here's one quick way to tell: Ask how many open source maintainers would be happy if someone else were to clone their open source project, rename it, claim that they had invented it, and have that clone completely overshadow and eradicate their original creation?

If the goal was purely altruistic, the original creator wouldn't mind. More candy in the infinite candy bowl, right?

But, in practice, many open source maintainers strongly oppose that. There is a strong culture of attribution in open source, largely because there is a compensation scheme built into creating free software: prestige. One of the main incentives that encourages maintainers to slave away day after day is the social cachet of being known as the cool person who made this popular thing.

> Imagine if this was the physical world, and you had a machine that could end world hunger. You could copy food like you could copy and paste information on a computer.

Analogies are generally bad tools for real understanding, but let's go with this. Let's say this machine took fifty years of someone's life to invent, toiling away in obscurity. Basically, an entire working career spent only on this invention with nothing else to show for their adult life.

If, at the end, no one would ever know it was you who invented it, how many people would be willing to sequester themselves in that dark laboratory and make that sacrifice?

Re: Open-sourcing AudioCraft: Generative AI for audio

#272

Post by emporas »

chrisjj wrote: post #218
emporas wrote: post #94

Earlier quoted context omitted.

If you mean to give a source of melody of 30sec and extend that melody into a full song, yes MusicGen can do that. There are two ways to extend a song based on a melody: 1) give a sample, and continue the song from that sample as close as possible, and 2) give a melody as an inspiration. They both work in varying degrees of success. Audiocraft on github, issues or discussion sections have a lot of questions answered.

Evidence? None of the demos suggest that is true.

Did they change the base model? If not, then the audiocraft_plus, is based on audiocraft, and it creates music of length close to 5 minutes.

I don't know if audiocraft_plus incorporates all three modalities of the release, MusicGen, AudioGen, and EnCodec. It uses MusicGen for sure, all four models, small, medium, large and melody.

https://github.com/GrandaddyShmax/audiocraft_plus

Re: Open-sourcing AudioCraft: Generative AI for audio

#273

Post by deepvibrations »

agilob wrote: post #199
gavman wrote: post #13

> MusicGen, which was trained with Meta-owned and specifically licensed music, generates music from text-based user inputs, while AudioGen, which was trained on public sound effects, generates audio from text-based user inputs. Meta is really clearly trying to differentiate themselves from OpenAI here. Open source + driving home "we don't use data we haven't paid for / don't own".

Goddamn, Facebook being the good guy...

Nah, this is just the modern tech playbook: First you open source stuff, then you can monitor all the related development happening and whenever you see areas of interest/popularity, you simply clone the functionality or buy out whatever entity is building that interesting stuff.

Re: Open-sourcing AudioCraft: Generative AI for audio

#274

Post by squidsoup »

famouswaffles wrote: post #122
randcraw wrote: post #115

Earlier quoted context omitted.

With 90% of human generated media content being forgettable within weeks of publication, and AI not yet capable of matching even average human content (much less pro level), it’ll be some time before we have to worry about AI overwhelming most media content and erasing the works of memorable human authors.

>and AI not yet capable of matching even average human content (much less pro level) Yeah this is not true. Sota Text, Image generation is well above average baselines. You can certainly generate professional level art on Midjourney

Commercial art and Art are not the same thing.

Re: Open-sourcing AudioCraft: Generative AI for audio

#275

Post by kmeisthax »

pbhjpbhj wrote: post #268
elondaits wrote: post #184

Earlier quoted context omitted.

The issue with “non commercial” is that no, it’s not well established. Licenses with a NC clause are so problematic to be practically useless. If you just want to use something at home privately you don’t need a CC license… a CC license is for use and redistribution. http://esr.ibiblio.org/?p=4559

>If you just want to use something at home privately you don’t need a CC license… // I presume you mean in USA, because in UK you don't have a general private right to copy. Our "Fair Dealing" is super restrictive compared to Fair Use.

Funnily enough in the UK they actually tried to fix this. The music industry argued that the lack of a private copying levy made legalized CD ripping into government confiscation of copyright ownership... somehow. The UK courts bought this, so now the UK government is constitutionally mandated to ban CD ripping, which is absolutely stupid.

Re: Open-sourcing AudioCraft: Generative AI for audio

#276

Post by crazygringo »

oceanplexian wrote: post #251
crazygringo wrote: post #131

Earlier quoted context omitted.

If one "large player" like the NYT decides to "alter the past", you can compare with the WaPo or any other newspaper. You can compare with the Internet Archive. You can compare with microfiche. These aren't "impossible to detect", they're trivial to detect if you bother to compare. We have tons of credible archived sources owned by different institutions. And these sources are successful in large part due to their cr…

> If one "large player" like the NYT decides to "alter the past", you can compare with the WaPo or any other newspaper. You can compare with the Internet Archive. You can compare with microfiche. These aren't "impossible to detect", they're trivial to detect if you bother to compare. Detection doesn't really matter, because people are too lazy to validate the facts, and reporters are not interested in reporting them.…

> and reporters are not interested in reporting them

You really think that if the NYT started altering its past stories, other publications would just... ignore it?

It would be a front-page scandal that the WaPo would be delighted to report on. As well as a hundred other news publications.

Thankfully.

Re: Open-sourcing AudioCraft: Generative AI for audio

#277

Post by chaxor »

Filligree wrote: post #205
voz_ wrote: post #192

Earlier quoted context omitted.

Can you chill? It’s def open source

So I can build a business on it, then?

Research does exist you know. This is immensely helpful for a huge number of people in academia.

If you want to build a company, perhaps you should do what everyone in the industry has done for millennia, copy the movements performed and optimize them while doing so.

Re: Open-sourcing AudioCraft: Generative AI for audio

#278

Post by karencarits »

cpill wrote: post #202
naillo wrote: post #195

Earlier quoted context omitted.

> Systems that don't allow people to extract value from the hard work they put into collective advancement [...] one's going to spend all day making candy and put it in the "free candy" bowl On the other hand this is what researchers do all day every day. PhDs and professors work for the common good and get barely any pay in return. Maybe the future model in art and music is more like the academic researcher.

yeah, but they also get to work on what they love as opposed to what ever the corporate interest currently is. it's rare you get paid will for doing what you love i.e. music, teaching, designing hand bags etc

Well, researchers usually have to get their own grants and must thus work on whatever various funding sources deem worthy. Further, academic positions typically have duties that researchers may not like - administration, reporting, teaching, etc

Re: Open-sourcing AudioCraft: Generative AI for audio

#279

Post by hobofan »

smallerfish wrote: post #124
CharlesW wrote: post #90

Earlier quoted context omitted.

Although I think AI-generated and AI-augmented (using voice cloning, etc.) artists are a given, for Spotify to stop paying labels they'd have to be able to remove all non-Spotify content from their streaming catalog. That doesn't seem like a possibility in our lifetimes. (Also, Spotify hasn't even been able to build a sustainable business on podcasts, which they copy to their closed platform for free.) It's an intere…

Their current pay structure is royalties, i.e. per listen. If they can route their audience to mostly AI generated content in time (say, 5-10 year transition), and it's just as good for most people, then they can negotiate much lower prices with the labels. We all grumble about Netflix being full of junk, but most of us are still subscribers, despite a sparse catalog of big name movies.

> then they can negotiate much lower prices with the labels

Or alternatively, if the labels are not stupid, they'll negotiate for a higher price per listen (or similar), as they are still as essential to the service as before.

Post reply on HN