Zhizheng Wu

A novel speech tokenizer with an end-to-end diffusion autoencoder and text-aware decoding, operating at 6.25 Hz and 0.0875 kbps

We propose a foundation model for unified speech generation with masked generative pre-training.