Tag
Proposes S2ST-Omni 2, a many-to-one compositional speech-to-speech translation framework that replaces flat language labels with structured typological priors to improve multilingual adaptation, achieving superior performance on CVSS-C.