Tag
Introduces Instruct-FD, a benchmark for evaluating whether full-duplex speech systems can follow explicit turn-taking instructions. Results show the best model achieves only 64.4% adherence, highlighting a significant gap in instruction-following turn management.