As an old fart, I can chime in for any younger fans on here that VocaTone has long had a reputation for being dickish, up to and including what @DefiantKitsune has said - and I can back this up, because I remember VT posting on VocaloidOtaku when I was a member there, and they eventually managed to alienate a good chunk, if not the entirety of the VO community with their messiness.
I find it funny to see this as a big deal when that's like... a big chunk of the most popular japanese vocaloid producers.
So I thought the art looked familiar and then I realized it's the same artist for Oliver's M.AI illustration (with some Oliver artwork in the doc). And Oliver's official age is 12. Oh boy.......
Awful. Originally I was going to be like "oh shotacon allegations are kind of the least most pressing matter in regards to M.AI" but....yeesh
As it's understandably one of the most contentious parts of this controversy, I've compiled all the statements regarding the use of RVC, Melodyne, and Synthesizer V in the creation of a Maghni AI demo for posterity.
The original screenshot of the Maghni devs discussing enhancing the demo behind the scenes, the authenticity of which has not been disputed by any party:
In this screenshot Suzie (demo artist) suggests that they splice in phonemes to improve the demo. Cocoa (Crescendia developer) agrees, while acknowledging that it is unethical. Editing individual phonemes in Melodyne is also discussed. Suzie later clarifies that she used Synthesizer V output, then converted it using RVC to create the spliced samples.
In her Twitter thread about the screenshot, Suzie says that consonants and a spoken voice line were generated with RVC (RVC is an open source AI voice changer, not associated with Maghni).
Austin Grissom's statement corroborates the use of samples from other engines, and adds that a Synthesizer V SVP file was used for the timing. He states that audio output from Synthesizer V was not one of the places they spliced from.
Natalia confirms that the demos were heavily processed, and corroborates that RVC was used. She confirms that RVC and 'related processing' were used for selection of consonants, and the spoken word portion.
And finally, Cocoa confirms again that consonants were spliced and RVC was used to create the spoken word section, and that Synthesizer V used as an editor. However, he disputes Synthesizer V being used for phonetic timing.
Suzie contradicts Cocoa and Austin to say that audio output from Synthesizer V was used by her to produce audio samples which were then fed into RVC.
All these stories, with the exception of exactly how SynthV was used, reinforce each other, and are consistent with Suzie's first statement. It's extremely clear what happened: I assume any confusion on social media comes from people skim-reading statements, or jumping to conclusions based on vaguetweets and secondhand summaries.
What's worrying is that Suzie's demo doesn't seem to have been the only misleading demonstration.
Cocoa and Suzie's final conversation implies that consonants from RVC or real singing audio were spliced into multiple demos. (Suzie was in charge of splicing on Eggtan's demo, whereas Cocoa recalls his own separate splicing tasks)
I stand by my initial opinion that this usage was unethical (especially if audio output from Synthesizer V was used in the process, as Suzie alleges), a sentiment I know the Maghni AI devs publicly and privately concur with (ahah). Now the spotlight is on them, I hope they'll be extra rigorous about showing accurate demos. I appreciate that they're now issuing apologies, but this is one incident that's best disproved with actions.
