ResearchBenchmarksAutomatic Speech Recognition on 800-hour curated dataset (test)Automatic Speech Recognition on 800-hour curated dataset (test)Follow1.62WERNV-ASR1.58881.79942.012.2206Jul 2, 2026Evaluation resultsMethodMethodLinksWERNV-ASRparameters=200M, archi...parameters=200M, architecture=RNN-Transducer, encoder=Emformer2026.071.62Whisper-Dparameters=1.55B, back...parameters=1.55B, backbone=Whisper-v2-large2026.072.4