Conversation doesn’t wait its turn: Sommelier, a data pipeline for real-time conversational voice AI

NAVER Cloud paper on Sommelier, a scalable open multi-turn audio pre-processing pipeline for full-duplex speech language models, presented at ACL 2026.