Metraipolis › archive › dc_78ccaf8689eb
Steering Follows Geometry, Not Labels: Emotion Directions in a Full-Duplex Speech Model — arXiv:2610.08887
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.08887
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-08T11:26:01.311Z
- SHA-256 of the archived text
c6b780e7bcf8ede7b477c7254e4794b30afd7cff1bf785e2c2de8a3269fc6391- Stable URL of this page
https://metraipolis.com/documento/dc_78ccaf8689eb
How to cite
Steering Follows Geometry, Not Labels: Emotion Directions in a Full-Duplex Speech Model — arXiv:2610.08887 — arXiv (en), https://arxiv.org/abs/2610.08887 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_78ccaf8689eb (Metraipolis archive dc_78ccaf8689eb, sha256 c6b780e7bcf8ede7b477c7254e4794b30afd7cff1bf785e2c2de8a3269fc6391)
BibTeX
@misc{dc78ccaf8689eb,
title = {Steering Follows Geometry, Not Labels: Emotion Directions in a Full-Duplex Speech Model — arXiv:2610.08887},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.08887},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_78ccaf8689eb; sha256 c6b780e7bcf8ede7b477c7254e4794b30afd7cff1bf785e2c2de8a3269fc6391},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.08887. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
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 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 only a few vector additions 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. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno). Categorie arXiv: Computation and Language (cs.CL) ; Sound (cs.SD). Pagina del lavoro: https://arxiv.org/abs/2610.08887 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-08T10:24:06.307Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).