Voicebank synthesis method which is based on a pre-trained voice model.
Distributed on the DiffSinger & etc. engine.