Which i believe is also referred to with the “create a custom vocalist based on an audio recording” in the spec ( see my first post in this thread ).
I’d agree with GovernorSilver, all the examples heard in the BJ video, had some level of a vocoder quality to them, some obviously with intent, and others considerably less so. This isn’t likely ever going to be “Deep Fake” audio quality vocals, where you read a series of sentences to model your speech with AI. Even those computation intensive methods have difficulty with singing. ( Reference : A radio show broadcast yesterday on BBC World Service. )
But on the other hand the S1000V is capable of a pretty great range of pitched vocalization, with just the turn of a dial. ( Can this be modulated ? Probably. ) Plus you have polyphony with the voice.
I think that avoiding actual language, and using this more for vocal “acrobatics” and effects will make the vocoder sound even less noticeable.
There are certainly some fun oddball features with this synth.
ADDED : Thank you GovernorSilver for the link to the manual.