1 paper · 1 filter
Feiyan Zhou, Luyuan Wang, Shoufa Chen +6
Modern audio generation predominantly relies on latent-space compression, introducing additional complexity and potential information loss. In this work, we challenge this paradigm…