Interactive Art

The Observer: an AI That Answers a Whole Room, Not One Microphone

C-LAB Taiwan Sound Lab's ongoing dance-performance project feeds spoken audience voices into a real-time system that replies in a cloned voice and reshapes live SuperCollider music around the mood of the conversation.

Most AI-performance systems are built around a single performer talking to a single microphone. The Observer, an ongoing project from C-LAB Taiwan Sound Lab, was presented at IRCAM Forum Workshops Hors-les-Murs Tokyo 2026 with a different premise: the audience itself is the input, and the room — not a performer at a mic stand — is what the system listens to.

Zone microphones instead of a single mic

Rather than requiring anyone to approach a fixed microphone, zone microphones listen to the whole room, so a single voice, a relay of people talking in sequence, a crowd speaking together, or even a shared silence can all be picked up and responded to as a group. Spoken language — in whatever language it was spoken — is fed into an AI system in real time, which answers within seconds in a cloned human voice, generating live dialogue alongside background textures and music described as “filled with reminiscence and deliberate distortion.”

A musician performs alongside the system, and each of their phrases is answered with the AI’s own sound in turn — a call-and-response structure where the human player isn’t accompanying the AI so much as holding a conversation with it in real time.

Ten sounds, synthesized live

All of the project’s audio — the cloned-voice responses, the ambient textures, the music that shifts with the conversation’s mood — is synthesized live in SuperCollider, moving between ten distinct sound states rather than triggering pre-rendered clips. That live-synthesis constraint is part of what makes the system responsive to a room full of unpredictable human speech rather than a fixed script: nothing about what the audience says is scripted, so nothing about the response can be either.

The IRCAM Forum presentation frames the project honestly as unfinished work — the talk covers how the system is built and what it does in a room, but also “what is still unresolved on the way to choreographing bodies,” meaning the eventual goal of extending this room-scale listening system into full dance choreography is still an open problem, not a solved one.