READ method uses acoustic discrepancy for reference-free ASR evaluation

Arxiv CS.SD (Sound/Audio ML)

The READ (Reference-free Hypothesis Evaluation with Acoustic Discrepancy) method evaluates automatic speech recognition hypotheses without reference transcriptions. It uses acoustic discrepancy to estimate recognition quality.

speech-to-textspeech-recognition

Arxiv CS.SD (Sound/Audio ML)·Zhihan LiHankun WangYiwei GuoBohan LiXie ChenKai Yu

7 days ago