VITS

VITS is an end-to-end text-to-speech model that generates raw audio waveforms directly from text, without a separate acoustic model and vocoder.

From the paper Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech by Jaehyeon Kim, Jungil Kong, and Juhee Son.

It combines a conditional Variational Auto-Encoder with normalising flows and adversarial training, similar to a Generative Adversarial Network. It also uses a stochastic duration predictor, so the same text can be spoken with different rhythms.

VITS is the basis of voice conversion projects like So-VITS-SVC.