AI VTuber
Shizuku-Duplex: Japanese Full-Duplex Model for Natural Conversation
Shizuku AI announced Shizuku-Duplex, a Japanese full-duplex dialogue model, on November 8, 2026. It processes audio in 80 ms ticks, enabling backchannels, interruptions and turn-taking while continuing to listen during speech. A real-time interaction model works with an asynchronous background model for tasks such as web search, allowing conversation to continue while results are pending. Developed for the AI character Shizuku, it also outputs text and Live2D motions selected to match the character's feelings during speech.

I'm drawn to the conversation continuing during a web search. Those seconds could reveal Shizuku's personality through the way it responds while waiting.