PulseAugur
中
实时 08:11:33
English(EN) GLARE: Generating Listening Heads with Appropriate Reactions

GLARE系统在对话中生成逼真的倾听者反应

研究人员开发了GLARE,一个用于在双向对话中生成逼真倾听者反应的新系统。现有方法在自然倾听者行为方面存在困难,缺乏恰当的反应标注和评估指标。GLARE通过引入一个包含147小时以上配对说话者-倾听者视频和跨六个类别的64,557个反应标注的新数据集来解决这个问题。该系统利用一个以Qwen2-Audio韵律和时间反应损失为条件的流匹配Transformer。一种新颖的评估协议测量反应发生、时间对齐和视觉质量,证明了GLARE在视觉保真度和行为逼真度方面均优于先前的方法。 AI

影响 这项研究可能带来更自然、更具吸引力的AI驱动的对话代理和虚拟助手。

排序理由 该集群包含一篇研究论文,详细介绍了一个用于在对话中生成逼真倾听者反应的新系统和数据集。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CV 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →

GLARE系统在对话中生成逼真的倾听者反应

本文如何被排名

Signal score
18 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
该集群包含一篇研究论文,详细介绍了一个用于在对话中生成逼真倾听者反应的新系统和数据集。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
paper, other
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

  1. arXiv cs.CV TIER_1 English(EN) · Zikai Liao, Yumin Suh, Yi Ouyang, Yi-Lun Lee, Yi-Hsuan Tsai, Zhaozheng Yin ·

    GLARE:生成具有恰当反应的聆听头部

    arXiv:2609.40317v1 Announce Type: new Abstract: While talking head generation has advanced rapidly, generating natural listener behavior in dyadic conversations, which know when to react, how to react, and with what type of response, remains underexplored. Existing dyadic dataset…