Researcher, Kuaishou- 快手科技
1 paper at NeurIPS 2025
We propose a foundation model for unified speech generation with masked generative pre-training.