Conversation Is a Two-Body Problem: Dyadic Evaluation of Full-Duplex Dialogue Models

Key points

• These single-sided frameworks evaluate only half of a two-body problem, where turn-taking, overlap, and interruption are joint products of two coupled speakers.1
• We propose DyaFDB, a framework that evaluates full-duplex models in a dyadic setup: two models converse directly under assigned roles with cooperative or conflicting goals, and both sides are scored offline with an external judge.1
• We instantiate four tasks as 140 scenarios and record 7,560 conversations, covering six self- and cross-play pairings.1
• We will release the scenarios, role prompts, and recording protocols between two full-duplex models, without any pre-recorded audio.1

Sources (1)

• [1]Conversation Is a Two-Body Problem: Dyadic Evaluation of Full-Duplex Dialogue Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 10:44 AM

  Full-duplex spoken dialogue models listen and speak at the same time, enabling voice agents to have natural, low-latency interactions that turn-based systems cannot offer.

  These single-sided frameworks evaluate only half of a two-body problem, where turn-taking, overlap, and interruption are joint products of two coupled speakers.

Extractive summary: sentences quoted from the sources.