Benchmark RadarAI BENCHMARK PROFILE
TurnBench
Evaluates end-of-turn and interruption detection in dyadic human conversation across six interaction styles using a 30-hour hand-labeled corpus and standardized metrics.
- Released
- 2026-08-25
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Provides the first multi-domain, triple-annotated corpus for turn-taking, enabling consistent comparison across conversation types and revealing type-dependent interruption errors.
Motivation
Speakers in natural conversation take turns speaking and listening, deciding in real time when to take, hold, or yield the floor.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.