Steering Follows Geometry, Not Labels: Emotion Directions in a Full-Duplex Speech Model

Full-duplex voice agents need to modulate emotion and delivery during real-time conversations, when de-escalating a complaint, carrying urgency in dispatch, softening a clinical result. Emotion and delivery control is well studied for TTS and turn based models through prompt-conditioned synthesis, reference-conditioned synthesis and activation steering; PersonaPlex~[10] controls identity in a duplex model but not affect. We study emotion steering in Moshi, a fully open sourced full-duplex speech language model, across four emotions, using mean-difference activation steering, which costs one vector addition per frame and no retraining. We show that emotion is linearly decodable from Moshi's residual stream, but activation steering is only partially achievable, and unevenly so; as happy, angry and surprise steer towards a shared direction while sad is distinctly steerable. We also show that the shared component across the three emotions cannot simply be projected away from all the emotions equally.