Shizuku-Duplex is a full-duplex audio model built for AI characters. It handles simultaneous listening and speaking with natural pacing, background tool use, and synchronized avatar actions.