Roxette - Feat - Marie Fredriksson Vocal AI - Sunflower

Roxette - Feat - Marie Fredriksson Vocal AI - Jezebel

Roxette - Feat - Marie Fredriksson Vocal AI - Walking on Air

Roxette - Feat - Marie Fredriksson Vocal AI - The Craziest Thing

7 songs have been completed from Marie Fredriksson Aka Roxette trained vocal dataset.
Marie was not available to record these songs with Per her band colleague due to her passing
Vocal AI Technology puts Marie back with Per in Roxette.

RVC2 Nvidia1600 was used locally to train Maries Vocals
Maries Trained vocal pack was then used used within www.tryreplay.io
weights packs for replay are Available free from from www.weights.gg

 

Can you use Replay to clone a guitar tone? How would you go about it? Are there better applications for that? Also, if you sing in English with an accent, will the cloning process smooth it out? Are the voice models language-dependent, like only English for now, or can I sing in Japanese with Ella Fitzgerald's voice? Thank you and sorry for the silly questions!

 

That's why I'd like to know as well.
A member here posted examples of his own guitar cloned to AI trained cello and sounded great. But I can't find the thread.
I'm guessing he used a vocal AI but not sure.

 

That's interesting, I'd like to know if there are AI that can separate orchestral instruments from the mix

 

Hi, not sur about the guitar tone (although some people made "voice" model with tooth brush or game noise, but it's not the same), maybe it could make an interesting tone with different articulation. Replay only use the voice as the source so I'm not sure if you would be able to change a guitar take into another guitare tone (but some plugin could probably do that, but I'm not expert in guitare setting).

For the accent, it will not totally correct the original source accent, but it can maybe improve it a little. I use a lot of English voice model on French source (and sometime the other way around) and they all keep the French pronociation.

But maybe a combinaison of tool could fix that, like using the original voice with a program ACE Studio to build the voice (using your original source for the note) and then using that result with Replay to choose the voice model. With a good trained model, it can correct those little AI artefact that ACE Studio can produce, and make it sound more natural. That's a lot of step, so maybe keeping the accent can be a better solution (especially if you're not trying to make a AI fake cover)

 

Slightly offtopic: Thank god to whoever brought this thread up in the last few days.
I've been looking for this exact program yesterday but couldn't remember the name of it, Google didn't turn up anything useful in the meantime.
Desperately need to train or find Paul McCartneys current voice for a little side project haha

 

I was looking for a Paul McCartney earlier tonight and found more than several from different time periods. Have fun.

Try to imitate him when you sing and it will sound more like him. I was hoping for James Earl Jones but the only one of him sounds bad.

 

Last edited: Apr 22, 2024