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

141–150 of 335 posts

Re: Open-sourcing AudioCraft: Generative AI for audio

#141

Post by notmypenguin »

cooper_ganglia wrote: post #82
gilmore606 wrote: post #72

Earlier quoted context omitted.

> I can't wait for it to happen, it's a low-cost way of producing even more music to listen to. I can't really understand this. I'm a DJ and a huge music nerd, and I spend a lot of time every week discovering new music from the past 100 years and all over the world, and I'm constantly struck by _how much of it there is_. I've spent weeks just digging through psych-funk records from West Africa from the 1970s. How can…

Music is self-expression. I don’t always identify entirely with others. I always identify with my self. Having music generated for you on such a personalized level is an attractive prospect. I don’t think this replaces “100% organic, human-made” music, though. I think there’ll always be a reason to listen to music made by other people. But I think this changes the landscape of how and why people create music to begin…

I don’t agree with your definition of music. For me, as both a musician and a listener, music is communication between human beings via harmonic carrier waves. Using a machine to make word salad copies of existing communiques is literally just nonsense to me

Re: Open-sourcing AudioCraft: Generative AI for audio

#142

Post by rvnx »

mcbuilder wrote: post #97
ipaddr wrote: post #55

Earlier quoted context omitted.

Checking OpenAI. Google is still playing checkers.

I just can't get how bad Google is doing. They have a ton of top researchers, papers, money, just no good LLMs. It's like OpenAI was first to the punch, and everyone else just saw $$$. Meta was smart to go down this open source road, as the masses will start training their llamas one way or another. Personally I believe the "intelligence" aspect will asymptote, so even having exclusive access to a "super AI" (i.e. hy…

The problem also is that Google is making lot of grandiose announcements about tools and models that nobody can see nor use. This is a serious credibility problem in the long-term.

Re: Open-sourcing AudioCraft: Generative AI for audio

#143

Post by praveenhm »

aedocw wrote: post #98
praveenhm wrote: post #87

does this model help in TTS(text-to-speech), badly needed only free option is bark and tortise TTS right now.

Coqui-TTS with vtck/vits is very good right now. Not as good as eleven labs or coqui studio, but for fast open TTS it's pretty good, in case you're not familiar with it. It will be great when there's eventually something open that competes with the closed models out there.

Excellent, I will take look into this.

Re: Open-sourcing AudioCraft: Generative AI for audio

#146

Post by frognumber »

hackernewds wrote: post #137
RobotToaster wrote: post #93

CC-BY-NC Isn't an open source licence, it violates point six of the open source definition https://opensource.org/osd/

who gets to declare what is the "open source definition" and why?

In my opinion, the Free Software Foundation, ironically, since they invented the movement, with open source starting out as a tacky rip-off with the ethics stripped out. After decades, open source converged on free software.

More popular opinion is OSI: https://en.wikipedia.org/wiki/Open_Source_Initiative

They were founded by the persons who (claimed to have) invented the term in order to steward it. It's the same definition as the FSF.

Re: Open-sourcing AudioCraft: Generative AI for audio

#147

Post by jononor »

crazygringo wrote: post #131
ysleepy wrote: post #126

Earlier quoted context omitted.

I don't agree. With ML tools it is possible to make sweeping changes to images and text that are often impossible to detect. combined with the centralisation of most online activities, large players could alter the past. Imagine facebook decides to subtly change every public post and comment to show some particular person or cause in a better light.

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…

The suggested large player was Facebook and Facebook posts. Which trustworthy independent sources of authenticity do we have for that? I do not think those you mention reach inside their walled garden?

Re: Open-sourcing AudioCraft: Generative AI for audio

#148

Post by klapinat0r »

s1k3s wrote: post #4

The demos are great. Could someone explain what’s in it for Meta open sourcing all these models?

Somewhat relevant, Yann LeCun insisted the research should be open sourced. At least in an academic sense.

He touches on it briefly in this podcast episode: https://www.therobotbrains.ai/who-is-yann-lecun

Re: Open-sourcing AudioCraft: Generative AI for audio

#149

Post by conductr »

ipaddr wrote: post #55
zyang wrote: post #44

Earlier quoted context omitted.

I was just thinking how Google made Android free to check Microsoft. This is Meta checking Google.

Checking OpenAI. Google is still playing checkers.

The fact you believe, rightly or wrongly, that meta is ahead of google on ai explains why meta would open source this. It’s a good reputation to maintain.

Re: Open-sourcing AudioCraft: Generative AI for audio

#150

Post by rvnx »

kmeisthax wrote: post #27
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".

This is purely a function of everyone remembering the RIAA's decade-long campaign to prevent people from taking the music they had rightfully stolen. As far as I'm aware LLaMA was trained on "publicly available data"[0], not "licensed data". Furthermore, MusicGen's weights are licensed CC-BY-NC, which is effectively a nonlicense as there is no noncommercial use you could make of an art generator[1]. This is not only…

Google is running on "publicly available data", not "licensed data"
Post reply on HN