Using AI it would be great to have celebrity voice packs. These would emulate the vocal stylings of popular celebrity singers.
For example:
Michael Jackson
Frank Sinatra
Christina Aguilera
Garth Brooks
Paul McCartney
The Beatles
John Lennon
Aretha Franklin
Ozzy Osbourne
Bon Jovi
Bruce Springsteen
2 Likes
I don’t like the idea, to be honest. But it almost certainly will happen.
We’ve already seen in the news people making unofficial and unauthorised AI vocal tracks using famous artists’ voices. It’s going to be impossible to stop, so the logical thing will be for some artists and record companies to license voices. Doing that will at least make them some money and give them some control.
2 Likes
Voice changers are one thing, but to make that sort of thing a reality for a commercial vocal synth is a different beast.
The training data used for development of SynthV voice databases is all appropriately licensed, so in order to have a celebrity voice provider you’d need to either hire them for studio time or pay a suitable license fee for a large number of high-quality recordings.
Frankly, vocal synths are not popular enough in the English market for this to be feasible.
Japan and China have much more established usage of vocal synths, which is probably a large part of why Weina exists and why we see many popular voice actresses as voice providers for the AHS voice databases (in keeping with the way Vocaloid voice providers were historically selected).
Eclipsed Sounds has already managed to work with some incredibly talented voice providers, but as far as celebrity talent goes… the price tag is probably well beyond the realm of possibility.
5 Likes
An interesting discussion topic for sure . . . . 
As the precise MO for database creation is (obviously) closely guarded with Kremlin-like security, as it IS of course Proprietary . . . . It could be that obtaining the necessary samples - not to mention the TRAINING DATA for Deceased Singers, from EXISTING recordings - may just be too steep a hill to climb . . . ?
As regards “celebrity” databases . . . . . WHO do you ask as to whether it’s commercially viable?
Whilst I’m firmly convinced that, as a rule - NOTHING should be allowed to interfere with the making of MONEY 
- This question could probably only be answered after a discussion between Dreamtonics and the “Celebrity” in question?
As for Vocal Synths popularity in the English Market - This would most certainly be improved if “English” - as opposed to “American-English” Databases were available - or at least - if it was simpler to De-Americanise the pronunciation?
I’m not sure it’s that closely guarded, and it’s certainly possible to guess what it looks like. For the sake of discussion, let’s assume it looks a lot like Kanru’s Arparsing reclist, only more complete. Each phoneme transition is represented multiple times, so there are multiple versions to select from.
For deceased singers, it’s unlikely that you’ll get sufficient coverage from existing material. And the quality of the recordings are likely to be all over the map. A singer’s voice changes over time - “young” Elvis sounds different than “old” Elvis, for example.
Something like trans-lingual training could be applied, so that a voice can be converted into another. That’s already been done with people using AI to apply “young” Paul McCartney’s voice to “old” McCartney recording, to make them sound more like his younger self.
The simplest thing to do might be to get a sound-alike vocalist to record the phonemes, and train the AI on the pitch information from existing recording.
You can generally find a right holder. For example, Elvis’ voice rights are likely held by the Authentic Brands Group.
The pronunciation is only American English in so far as the vocalist sings with an American English accent. See Ninezero for a case in point, or the “English” voices that still retain an Asian accent.
I’m sure as time goes on, there will be additional voices that aren’t American English. I wouldn’t mind an AVANNA-like voice.
Soapbox Time. (Feel free to stop reading here 
)
Personally, I don’t think using “celebrity” voices is a useful route to go down. For one thing, acquiring rights is an additional expense. I can’t imagine the rights holder allowing the use of a celebrity voice without requiring some percent of the royalties.
Plus, singers typically create brand value by selecting excellent songs during their lifetimes. By ceding quality control and allowing a celebrity voice to be used on songs, the rights holders would create competition against their existing catalog of music, and likely dilute the brand quality by allowing the market to be flooded with knock-offs.
Finally, I think there’s a coming backlash against the use of AI, and synthetic celebrity voices would likely be a focus of attention.
Elvis certainly wouldn’t think my songs are good enough to sing, although he might have done it if I gave him lots of money. 
I wouldn’t like it if someone did that to my voice.
I’m happy to have a “generic” synthetic vocalist sing the song, so it can be judged on its own merit.
4 Likes
It’s a fascinating topic for speculations about the future 
If a voice synth could faithfully reproduce even a couple of the artists u4eamusic lists in the opening post, I think demand in English-speaking countries [US mainly] would be enormous. And that might make it lucrative enough for licensing in some cases.
This being said, if I were Bruce Springsteen [for example], I would definitely not want my voice singing about things I didn’t support. Or for that matter singing on songs that were mostly not good. This is why I wouldn’t be keen on the idea…especially for dead artists.
Is something that will happen. I am almost certain, considering (1) is possible (2) is basically the first/second thing people think of when discovering this techonolgy (meaning there is huge public interest).
I havent tried myself, but considering the voice replacing techonology is already out there replacing other people’s voiced with that of famous artists, think you can already approximate that. For example, create your song with NineZero and then replace it for Kurt Cobain’s. Or create with Asterian, then replace with Barry White’s… I guess the closer the replaced and replacing voices are, the better the final result.
Regarding proper, licensed Voice banks, as the technology grows, more famous people will put an eye on it. Only a matter of time until some celebrity decides to be a “pioneer”, gives even more publicity to it and this tech becomes thenextbigthing beyond China and Japan.
I am convinced it will happen in a few years time.
It will be an interesting event to see unfolding.
I can see even the heirs of deceased artists given permission for the development of voicebanks based on them.
Ethic issues will arise and legislation will be needed for certain cases such as those.