2026 03 Preprints

New preprint on speech: Sommelier — scalable, open multi-turn audio pre-processing for full-duplex speech language models.