Take a look at what Voctro Labs are up to these days. They also published some demos of some new software last year, which pretty much did just what you described: used AI to make a model of a voice from a small amount of recordings (maybe 40 minutes of speech data IIRC, which, well, I've made a free virtual singer, being able to record everything you need in 40 minutes is AWESOMELY efficient), then sang songs using that model and using vocals sung by another person as a controller. I think.wagtunes wrote:Okay, here it is. Because of my frustrations with Vocaloid (all the criticisms) this is the VST that I would pay $1,000 for if somebody would make it.
Here's how it works.
You sing into a mic. The audio then goes through the VST and converts YOUR voice into any singer's voice you want it to sound like and perfectly on key. So, you could sound like Sinatra, Bon Jovi, Meatloaf, Cher, anybody you can think of.
This is their current product: https://www.voiceful.io/
