A new arXiv preprint introduces MultiPartyTurnBench, a benchmark for evaluating floor control in multi-party spoken dialogue. The work targets full-duplex speech systems, which must decide when to enter, hold, or relinquish the conversational floor.

According to the abstract, existing evaluations largely study dyadic turn-taking or end-to-end performance; the sentence is cut off in the source, so the full comparison is not available. The benchmark appears aimed at the more complex setting where multiple speakers overlap or backchannel without taking the floor.

Floor control is a core practical problem for conversational AI, but it is less studied than two-person turn-taking. MultiPartyTurnBench seems intended to fill that gap, though details of its tasks, metrics, and results are not included in the abstract provided.