Read What You Hear: Reference-Free Hypotheses Evaluation with Acoustic Discrepancy
READ is a novel, reference-free metric for evaluating Automatic Speech Recognition (ASR) hypotheses directly from speech signals. It uses a pre-trained Text-to-Speech (TTS) model to calculate acoustic discrepancy by measuring the conditional likelihood of speech tokens given a text hypothesis. This method emphasizes acoustic grounding over linguistic evaluation, offering robust performance even under noisy conditions.