Our work on Continuous-Token Diffusion for Speaker-Referenced TTS in Multimodal LLMs appeared at ICASSP 2026.