See that doesn’t sound any different from just having them record a phonemic inventory for a traditional text-to-speech system, it’s just simplifying the process and making it so the text doesn’t need to have a corresponding pronunciation key, and presumably meshing it in a bit better.