ResearchBenchmarksAutomatic Speech Recognition on LibriSpeech (retain set)Follow0.06Word Error Rate (WER)Finetune-3.937623.046250.0377.0138Mar 4, 2025Evaluation ResultsMethodMethodLinksWord Error Rate (WER)Finetune2025.030.06Original2025.039.99OrthoGrad2025.0312.11GDR-GMA2025.0317.38NegGrad+2025.0372.87SCRUB2025.03100