For signing up for a free demo:
https://www.wandb.com/papers

 

So we will get new Elvis songs with his original voice, soon, as it seems. Of course, they'll have to train the AI to sing a "bit" too
But absolutely impressive so far, for sure!

 

It's really cool and scary at the same time. Is that tool could become a plugin or software we can easily use as professionel, it would be great, for exemple we could do a translated version of an actor with the original actor's voice (with the new iZotope thing to match the ambience), but also, it would open the door to fake song, like giving the possibility to everyone to have Adele or Kurt Cobain singing an original song you wrote. From a legal point of view, it's like a future nightmare, maybe we'll start to see copyright on voice texture (which open the door to copyright on sound design).

And I'm not even talk about all the fake news possibility (which will be a thing for sure, with the rise of deepfake video).

Lot of possibility, but also lot of future problems that could make this thing not available for everyone in an easy way to use, which is sad for us

still waiting for the R2R version to pop up

 

And the question is: Why do they want our data?

 

has anyone signed up? is this an app on their server or what? what's the program called?

also, they're saying the software creates a spectrogram and plays a spectrogram back with a vocoder so I think this is a red herring because 1) a spectrogram is meaningful only to the eye of a human being and 2) a vocoder does not synthesize realistic voice, only formant synthesizers do that.

 

The program you're referring to is likely a service hosted on a server, and it sounds like it involves AI-generated audio. The spectrogram and vocoder method might be a step in the process, but you're right that a spectrogram is visually meaningful and a vocoder alone doesn't create realistic voices—formant synthesizers are typically used for that. If you're wondering how to split stems, the software might use similar techniques for isolating and manipulating different audio elements. The exact program name and details should be clarified when signing up for the demo.

 

Wait is this tech 5 years old?

 

Is this something like dreamtonics vocoflex?

 

I would say yes. Vocoflex need about 10 seconds of voice to clone. This seems to be more refined.

 

I guess this is the way music is going. Both clever and sad simultaneously.

 

backup singers and session vocalists. there's another job down the drain.
and for the naysayers - wait until you tell wallmart that you can either book a 60 piece choir for 2K for the music of their commercial or we can use this plugin that emulates that perfectly for 200 bucks and we'll see about that "oh AI won't replace anyone"

 

Not really if you think about the actual singing, it's much more than tone and timbre

 

I've just cut spending with those last week, and I'm a beginner producer. (But it was A.I. in the same sense that any VSTi is A.I., it's still writing, composing, MIDI performing, CC programming, learning he tools etc. — not mere prompting.)