UDIO AI- short serial piece for baritone and orch

[jbuhler]

UNFORTUNATELY give it time and it will respond better to prompts. It will adjust to sync points, and the director's (prompter's) instructions about where to go intense or where to go more unnoticed, what to convey on this or that point etc etc....

Only if there is investment to make this to happen. It won't happen on its own. It's not clear to me that the tech bros understand the music market well enough to target it the way it would need to be targeted. They will also need to understand how to tag the data, which we all know is not a simple task that even companies that have been doing this a long time don't really excel at. And the market has to be big enough to warrant the investment. Right now these AI music engines seem focused on the popular music market, and that won't get them what they need for media music.

 

[Jett Hitt]

Other than the performance, I haven't personally heard anything that impressed me. The performances are frighteningly realistic, and I think that this is its ace in the hole. A stunning performance can make the most trivial piece of crap sound almost profound. What I see happening is the death of the media composer. Hollywood will leap at this. The director will sit there entering prompts until he finds a desirable selection that underscores his picture. No more paying composers to craft something that may or may not fit the bill. No more hiring orchestras to record the work. The past 20 years have seen film scores moving toward digital orchestras and unmemorable soundtracks. (I can't remember one theme from any of that Marvel crap.) First Spotify nuked all the album sales, and now the pay for the creation process is on the chopping block.

This may seem like a bleak prediction (and I hope I am totally wrong), but never underestimate the power of greed.

 

[SvenE]

I found this article somewhat helpful:

Audio Diffusion: Generative Music’s Secret Sauce

Exploring the principles behind diffusion technology and how it is being used to create groundbreaking AI tools for artists and producers.

towardsdatascience.com

Finally, a partial explanation. Very interesting article. Thank you for sharing.

 

[dcoscina]

I put in "Song in the style of Jerry Goldsmith's Planet of the Apes" and this is what I got. Very generic, meh. Thankfully

Udio | Apex of Dawn by BartokRulz29

Make your music

www.udio.com

 

[tebling]

I've posted this elsewhere, but since we're all putting our Nostradamus hats on I'll throw some "fun" predictions into the ring...

1. Assume the technology is "perfected" (likely in the next 3-5 years). By that I mean it follows some asymptotic curve where eventually its output is essentially indistinguishable from human made music. Maybe only 5% of the population can tell the difference. Remember, what we see today is just the beginning. To me it doesn't make sense to assess the impact based on its current state.

2. Spotify buys Udio or a competitor (it doesn't matter which). By this point the music streaming giant has accumulated enough user preference data and ML know-how to be able to accurately correlate a user's tastes with the tags that generate the same type of music. It uses this data to start creating personalized tracks, and introduces a "tracks made just for you" feature.

3. The music streaming industry as a whole pivots rapidly from a user experience based on music discovery to one based on music creation. Want a new Beatles* song that nobody else has ever heard - generated for you every day? No problem.

4. Artist payouts start to diminish as AI created music gains market share. A savvy Big Tech player acquires Spotify in order to capitalize on this growth and uses their legal muscle to bend copyright law to its will, much in the same way Disney has in the past.

* These services aren't so brazen (yet) as to allow for direct impersonation (yet), but I'm sure it'll be easy to get close enough to avoid lawsuits and still feel like the "real thing".

 

[StefanoM]

We are at the Jurassic age for this tech- and everything seems possible now

Absolutely, but I'm living in the here and now.

What the future holds is uncertain; it could bring anything or its complete opposite. Technology doesn't always progress in the expected ways, and in this instance, it's not merely about technology but also numerous overlapping social issues.

Therefore, making any predictions is tricky—they tend to end up as mere speculation. Take, for example, the enormous amounts of energy these systems need to operate.

That's just one of many issues.

The challenges of widespread usage are vast, covering ethical, social, and technical aspects. The impact they have—and could potentially have—spans various sectors and virtually affects EVERYTHING. Thus, we're dealing with a phenomenon that's tough to forecast.

At the moment, some people are just surfing the trend, trying to make a quick profit, especially given the significant investments required to kickstart this entire setup. But once more, I don't have a clear vision of how things will unfold. That’s why I focus on the present, or at most, the immediate future.

 

[dcoscina]

Other than the performance, I haven't personally heard anything that impressed me. The performances are frighteningly realistic, and I think that this is its ace in the hole. A stunning performance can make the most trivial piece of crap sound almost profound. What I see happening is the death of the media composer. Hollywood will leap at this. The director will sit there entering prompts until he finds a desirable selection that underscores his picture. No more paying composers to craft something that may or may not fit the bill. No more hiring orchestras to record the work. The past 20 years have seen film scores moving toward digital orchestras and unmemorable soundtracks. (I can't remember one theme from any of that Marvel crap.) First Spotify nuked all the album sales, and now the pay for the creation process is on the chopping block.

This may seem like a bleak prediction (and I hope I am totally wrong), but never underestimate the power of greed.

I'm taking the Glass is half full stance on this. For composers wanting really stellar renderings of their hard work, I can see this working really well. It won't replace the other tools to compose, but having this quality of performance is something else. Even StaffPad cannot get this level of expression. I suspect DWH and MuseScore will be the first to exploit this resource if I had to place bets. Dorico and Sibelius might be hindered by legalities or budgets to apply this kind of tech but Ultimate Gtr apparently has a gazzillion dollars to put towards R&D.

 

[Brendon V]

I've posted this elsewhere, but since we're all putting our Nostradamus hats on I'll throw some "fun" predictions into the ring...

1. Assume the technology is "perfected" (likely in the next 3-5 years). By that I mean it follows some asymptotic curve where eventually its output is essentially indistinguishable from human made music. Maybe only 5% of the population can tell the difference. Remember, what we see today is just the beginning. To me it doesn't make sense to assess the impact based on its current state.

2. Spotify buys Udio or a competitor (it doesn't matter which). By this point the music streaming giant has accumulated enough user preference data and ML know-how to be able to accurately correlate a user's tastes with the tags that generate the same type of music. It uses this data to start creating personalized tracks, and introduces a "tracks made just for you" feature.

3. The music streaming industry as a whole pivots rapidly from a user experience based on music discovery to one based on music creation. Want a new Beatles* song that nobody else has ever heard - generated for you every day? No problem.

4. Artist payouts start to diminish as AI created music gains market share. A savvy Big Tech player acquires Spotify in order to capitalize on this growth and uses their legal muscle to bend copyright law to its will, much in the same way Disney has in the past.

* These services aren't so brazen (yet) as to allow for direct impersonation (yet), but I'm sure it'll be easy to get close enough to avoid lawsuits and still feel like the "real thing".

Just want to add to this, and say what if an artist like the Beatles licensed the rights to generate AI music using only their songs as training data. Or started their own music service doing this.

We have seen something similar already with Grimes, who launched her own AI software to simulate her voice and who encourages her fans to do it and splits the royalties 50/50 for any songs produced.

Suspect we will see more of this, but imagine if you create new Beatles generated songs from a model legally trained on their songs. And it provides a legitimate revenue stream to the original artist.

(I guess it also begs the question whether music labels could create AI versions of songs from bands who are no longer together or dead from just owning the rights to their back catalog, a scary thought that maybe bypasses copyright restrictions)

 

[jbuhler]

Just want to add to this, and say what if an artist like the Beatles licensed the rights to generate AI music using only their songs as training data. Or started their own music service doing this.

We have seen something similar already with Grimes, who launched her own AI software to simulate her voice and who encourages her fans to do it and splits the royalties 50/50 for any songs produced.

Suspect we will see more of this, but imagine if you create new Beatles generated songs from a model legally trained on their songs. And it provides a legitimate revenue stream to the original artist.

(I guess it also begs the question whether music labels could create AI versions of songs from bands who are no longer together or dead from just owning the rights to their back catalog, a scary thought that maybe bypasses copyright restrictions)

Licensing is the obvious work around to overt copyright infringement, though it’s not clear to me if the training data of the base model is submerged enough that it would never risk infringement of nonlicensed music in the base training data. It will be hard to use commercially with a risk of a copyright claim.