Continue, Adapt, or Yield: In-Turn Adaptation to Overlapping Speech in Full-Duplex Agents

A new evaluation called Duplex Cue finds that full-duplex voice agents adapt to listener contributions mid-turn far less often than humans, according to a single-model case study of 300 human-confirmed cues from unscripted English conversations. On the 66 collaborative pairs retained, recorded human speakers adapted in 68.2% of cases versus 34.8% for PersonaPlex, which otherwise continued unchanged (42.4%) or yielded (22.7%). The work argues that evaluating natural voice interaction requires measuring in-turn adaptation, not just whether an agent keeps speaking or stops.

Collection Weekly signals in speech-to-speech and full-duplex voice AI. Latest: 2026-W35, Aug 17 - Aug 23, 2026. Archive: fullduplex.ai/signals • 24 items • Updated /collections/otoearth/fullduplex-signals Papers /papers Continue, Adapt, or Yield: In-Turn Adaptation to Overlapping Speech in Full-Duplex Agents Abstract Duplex Cue evaluates how voice agents adapt to listener contributions during ongoing turns, revealing that current models adapt far less often than humans. thinkingmachines/Inkling-Small /thinkingmachines/Inkling-Small Full-duplex /papers?q=Full-duplex evaluation often emphasizes whether an agent keeps speaking or stops. That binary cannot express a third response humans use routinely: continuing to speak while incorporating what the listener just contributed. The contribution may be a missing word, a correction or a clarification. We introduce Duplex Cue, an evaluation of this in-turn adaptation /papers?q=in-turn%20adaptation in full-duplex /papers?q=full-duplex voice agents. Duplex Cue separates listener intent backchannel /papers?q=backchannel , collaboration /papers?q=collaboration , or interruption /papers?q=interruption from speaker behavior: continuing unchanged, adapting within the turn, or yielding. Adaptation includes acknowledgment as well as content revision. In a single-model case study using 300 human-confirmed cues from unscripted English conversations, we compare recorded human responses with PersonaPlex /papers?q=PersonaPlex continuations generated while replaying the listener's audio. We retain 208 pairs with the ongoing speaker active at cue onset and a scorable response in each condition. On the 66 collaborative pairs, recorded speakers adapt in 68.2\% of cases, compared with 34.8\% for PersonaPlex /papers?q=PersonaPlex . The model otherwise continues unchanged 42.4\% or yields 22.7\% . These findings show why evaluating natural voice interaction requires measuring how an agent responds to a listener's contribution as well as whether it keeps speaking. Get this paper in your agent: hf papers read 2609.13117 Don't have the latest CLI? curl -LsSf https://hf.co/cli/install.sh | bash Models citing this paper 0 No model linking this paper Datasets citing this paper 0 No dataset linking this paper Spaces citing this paper 0 No Space linking this paper