CMU and Idiap Build Synth-DoPaCo, a 1,329-Hour Medical AI Benchmark

A new open dataset of 8,800 synthetic doctor-patient conversations with 1,329 hours of audio tests whether audio language models can summarize long clinical visits.