For all of the blind people who just lost access to the Sonos app, and everyone else who’s accessibility done gone away, I dedicate The Accessibility Blues.
@bryansmart This song is awesome!!! Love the lyrics. However, what really gets me is the stereo separation in this song, and that classic tavern feel. Nicely done.
@wmj1102 I think, for that one, I was in manual mode, and used "blues, electric delta blues, electric guitar, male vocalist". Then, you re-roll several times until you get something like you want. There are several ways of doing Electric Delta Blues, you know, and several types of male vocalists it could come up with, but I had a good idea in my imagination of what route I wanted to go. You get it started on the right path, and extensions follow that.
@wmj1102 The AI can come up with anything, of course, but you get better results if you can imagine as much of what you want it to be, before you start. Imagine the story. Imagine the band and the vocalist. The AI won't do exactly what you want, but, when it feels like it is doing something close, then run with that. Treat it more like a writing partner, rather than an employee.
@bryansmart Considering that Udio and Sonos trained their models by downloading and scraping untold millions of songs from the web without consent, credit or compensation to make a commercial product, I think this is in bad taste.
The only reason this sounds any good is because of the uncompensated labour of workers who are now being digitally resurrected as zombies to perform.
@Stealcase I'm sure I've heard that critique before, and seen other scolding remarks using this same style. Did you obtain proper consent before appropriating it from them?
@bryansmart Seems you're making a legal argument, but I'm happy to explain why the equivalence you're presenting doesn't hold up.
There's no protection of Ideas under copyright, only the expression of ideas.
AI models trained on expressive works (Copyrighted songs) copy the expression of those works, not the ideas.
You're using a machine that from the get-go is infringing by training, to simulate the expression of creative works of others.
@Stealcase Haha! Every musician ever is a biological machine that has been trained to simulate the expression of performances by other musicians. You were trained to write me these arguments by mimicking ones you previously read or heard. I don't owe Ray Charles's estate every time I sing a soul song, nor do I owe Holt and Mifflin every time I write an essay. For music, copyright covers lyrics, melody, and mechanical recording. That is all. Style has no legal protection.
@bryansmart You are conflating your own creative process and expression with that of a machine that is trained through repeated matrix multiplication to replicate sounds millions of times.
We are not stochastic parrots repeating training data, we are human beings with critical faculties who can thoughtfully engage with information and inspiration.
You are doing a disservice to yourself.
Also, these songs you generate through Udio and Sonos have no copyright protection.
@Stealcase The songs aren't stochastic. I wrote the lyrics. I chose the style and song structure. A machine simulated performers. Keyboards have been doing that for 40 years. They're just better now. I have no moral reservations about using a machine to work out my ideas. I also don't care about copyright for my creations. I own a retail business. Music is a fun thing I do, which I don't care about making any money from. If people laugh and enjoy it, then good.
The difference between keyboards, synths and other digital audio equipment is that they're not built on the uncompensated labour of millions of musicians, they don't require training data.
Apps like Udio are replacing the need to pay musicians, which results in less people being able to afford making music for a living.
But that results in less real training data, and eventual AI model decline. A technology snake eating it's own tail. A black hole, an ouruborus.
@bryansmart The outputs of the machine are absolutely stochastic: they are a probabilistic average of the training data, using your inputs as a seed.
If I use the same inputs as you, and the same original seed, the output will be exactly the same. This is what stochastic means.
@bryansmart For what it's worth, I think your lyrics are great and its a funny idea. I just think how this tech is currently being built is unethical, and there are clear reasons why it shouldn't be used as-is.
But there are also pathways of building the tech in an ethical way, and since we are only on the precipice of adoption, we can dictate how we want this to go and how models should be trained. It's now or never.
@Stealcase Where we fundamentally disagree is I have contempt for copyright and intelectual protectionism. I have zero sympathy for someone demanding they get paid, over and over, any time someone uses something they made, long ago. Same for people who want to block progress, because they control the old way something works, and a new way threatens the control they've established. They all have personal interest in inhibiting progress to benefit just themselves.
@Stealcase I used to be a programmer and music production pro, so I know. The problem is we're living in an age where copying is easy as breathing. Copyright is a post-pay model, only possible with artificial restrictions. It's doomed to either fail, or overly oppress people. Before recorded music, creative work was commissioned/sponsored in a pre-pay model. That's all that makes sense anymore. Creatives aren't entitled to work as they'd prefer by oppressing others.
@bryansmart You're suggesting that creative workers "prefer oppressing others" simply by asking that their work not be used by companies with billion-dollar evaluations for profit.
This is a little unhinged.
I guess we should just submit to corporations who want to monopolize artmaking by commercializing subscription models trained on human culture. Commodify art, make sure AI companies are the only ones making money from art. Let them censor and control.
@Stealcase Well, that’s the darkest interpretation, but copyright oppresses by saying "You know that popular song everyone is singing? If you post yourself singing it, you’re banned," or "You know that popular story everyone likes? If you tell stories about those characters or places, I’ll sue you and take your house."
@bryansmart You are writing in present tense, but your examples are pure fantasy.
People release covers constantly on YouTube with no issue, the record label just gets the ad revenue (screw record labels by the way, there's nuance here).
You're allowed to sing whatever song you want in your home or in public. If you want to perform at a concert, it becomes more complicated.
Fanfiction has a deep and rich history online, its more popular than ever.
@Stealcase Anyway, we all enjoy it. Sorry your priesthood is threatened. I'd suggest you study up for a new vocation, if you don't want to starve. Best of luck to you.
@superblindman@bryansmart Speaking of Udio, have you heard the demos that Eleven Labs came out with showcasing their music tool to rival both Udio and Suno?
@bryansmart I was able to update my system today. The app scenes fine to me. It's just a whole different layout that you have to learn. But from what I can tell it's doable if your patient.
@InsaneAtheist Don't know. Others tell me basic swiping is broken. I don't have it, so don't know the details. Just sympathetic to overnight updates taking accessibility away.
@bryansmart I really like how unplugged this sounds. Maybe it's because the drums sometimes sound a little off beat, and the guitar in the right channel just sounds like some guy jamming live with his buddy. Real neat stuff!
Add comment