Describe the request
Grok Bot voice works well one-to-one. For office meetings and live research conversation it hits a hard limit: voice is one Bot, one caller, and unavailable in group chats (Voice chat help).
Ask: a multi-person / room voice mode that is speaker-aware:
- Preferentially treat one designated voice (e.g. the PI / account owner) as the interactive addresser.
- Do not treat other speakers in the room as direct asks — or at least separate them so the Bot can tell when it is not that person talking.
- Default behavior for ambient speech: keep a running transcript / notes; speak or act only when gated (wake phrase, named address, or clear address to the Bot).
Use case is scholarly practice, not a second product surface: a Bot in the room for colloquia and unit meetings that can scribe live research conversation and field questions when invited — closer in spirit to room-aware meeting assistants, but gated so overheard talk is not an instruction.
Beta offer: Embodied Cognitive Science Unit (OIST) can serve as a working testbed in real office meetings if there is a beta.
Happy to share failure modes from current one-on-one voice attempts (everything heard becomes an addressed turn) if useful to the team.