Mark Hasegawa-Johnson

GigaSpeechBench Exposes ASR Gaps Beyond Standard Benchmarks

A 680-hour human-annotated benchmark combines multilingual, dialect, accent, domain, and age stress tests across ASR and speech translation.