Published on August 28, 2026, TurnBench includes 30 hours of dyadic speech annotated by three people, a scoring protocol based on conversation analysis, and results for 14 systems.
TurnBench is an open benchmark for evaluating how systems handle turn-taking in spoken dialogue. Its material includes 30 hours of dyadic speech annotated by three people; evaluation uses a scoring protocol based on conversation analysis and reports results for 14 systems. The description highlights turn endings and backchannel cues such as “mm-hmm”.
Engineers can consult the benchmark to study these aspects, but the summary gives no scores and does not identify the systems evaluated. To verify the details, search for the original title, “TurnBench avalia a alternância de turnos em diálogos falados,” and check the published protocol, annotations, and results. If you use AI to analyze your organization’s materials, apply the appropriate data policy before submitting sensitive content.