New benchmark Uh-Mazing reveals disfluencies impact speech translation quality

A new benchmark called Uh-Mazing has been developed to study the impact of disfluencies in speech translation. Current systems often remove these speech irregularities, leading to a loss of meaning. The Uh-Mazing benchmark, which includes human translations and disfluency annotations for English speech translated into eight languages, reveals that false starts and self-repairs significantly affect translation quality. The research also demonstrates that these translation quality losses can be mitigated at inference time without the need for retraining. AI

IMPACT Highlights the importance of preserving speech disfluencies for more accurate and meaningful machine translation.

RANK_REASON The cluster contains a research paper detailing a new benchmark and findings related to speech translation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• English
• Hugging Face
• SpeechLLMs
• Uh-Mazing

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →