EgoSpeak: Learning When to Speak for Egocentric Conversational Agents in the Wild

Junhyeok Kim, Min Soo Kim, Jiwan Chung, Jungbin Cho, Jisoo Kim, Sungwoong Kim, Gyeongbo Sim, Youngjae Yu ♣ \clubsuit Yonsei University ♢ \diamondsuit Multimodal AI Lab., NC Research, NCSOFT Corporation · 2025

Read EgoSpeak: Learning When to Speak for Egocentric Conversational Agents in the Wild by Junhyeok Kim, Min Soo Kim, Jiwan Chung, Jungbin Cho, Jisoo Kim, Sungwoong Kim, Gyeongbo Sim, Youngjae Yu ♣ \clubsuit Yonsei University ♢ \diamondsuit Multimodal AI Lab., NC Research, NCSOFT Corporation (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-14892.txt.

Description

Junhyeok Kim and co-authors tackle a timing problem for wearable conversational agents in a 2025 paper. An agent that sees through its user's eyes must decide, moment by moment, whether to speak or stay silent. Speak at the wrong time and it annoys; stay silent and it misses its chance.

Their system, EgoSpeak, learns that judgment from first-person streaming video. It processes color video online and untrimmed, like a real conversation. To train at scale they built a pretraining set from in-the-wild YouTube conversations. On two benchmark sets the system beat random and silence-based baselines in real time, and richer context clearly helped.

Catalog record

CREATORJunhyeok Kim, Min Soo Kim, Jiwan Chung, Jungbin Cho, Jisoo Kim, Sungwoong Kim, Gyeongbo Sim, Youngjae Yu ♣ \clubsuit Yonsei University ♢ \diamondsuit Multimodal AI Lab., NC Research, NCSOFT Corporation
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-14892

Browse the full library · About the Tower of Babel