SpeechDx: A Multi-Task Benchmark for Clinical Speech AI

In three linesSpeechDx is a multi-task benchmark for clinical speech AI covering 12 datasets and 27 tasks across diverse health conditions. Tasks are structured by speech production stages (conceptualization, formulation, articulation). Evaluation of 12 audio encoders shows large-scale speech models outperform domain-specific ones, but none generalize reliably across clinical speech.

Summary generated by Claude — human-verified