What this workflow means
Singing clones and speech clones are different models
A speech clone is trained to read text convincingly. A singing clone has to hold a note, move between registers, and follow a melody that someone else performed. That is why Voxmin asks for sung training material rather than a voicemail-length speech sample, and why the model is applied to a source performance rather than to raw text. It converts vocal identity; it does not invent a melody for you.
Understand song generation vs voice conversion
