New VoxSumm corpus enables joint speech summarization and translation

Researchers have introduced VoxSumm, a new benchmark corpus designed for joint speech summarization and translation (JSumT). This corpus contains over 10,000 BBC article-summary pairs in 24 languages, totaling approximately 703 hours of spoken data. The goal of JSumT is to generate a concise summary in a target language directly from a long spoken document in a source language. Initial evaluations showed that Gemini-3.1 Pro performed consistently, and summarization into English was generally more effective than into other languages. AI

IMPACT Establishes a new benchmark for multilingual speech processing, potentially advancing cross-lingual summarization and translation capabilities.

RANK_REASON The cluster describes a new academic paper introducing a novel dataset and task for NLP research. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• British Broadcasting Corporation
• Gemini-3.1 Pro
• VoxSumm

• paper
• other

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