Learn2Chat: Rethinking Dyadic Talking Heads via Interaction-Modulated Monologic Priors
📄 Learn2Chat: Rethinking Dyadic Talking Heads via Interaction-Modulated Monologic Priors 标签:#音视频生成 #Transformer #生成模型 #自监督学习 #音频理解 6.8/10 | 创新 1.6/2 | 严谨 1.2/1.5 | 实验 0.9/1.5 | 清晰 1/1 | 影响 0.5/1.5 | 开源 0/1.5 | 复现 0.3/0.5 | 工程 1.3/1.5 ✅ 6.8/10 | 前50% | 文档类型:方法研究 | 评分置信度:高 | #音视频生成 | #Transformer | #生成模型 #自监督学习 | arxiv 👥 作者与机构 第一作者:Zikai Huang (South China University of Technology, School of Computer Science and Engineering) 通讯作者:Shengfeng He (Singapore Management University, School of Computing and Information Systems) 作者列表:Zikai Huang (South China University of Technology, School of Computer Science and Engineering), Siyue Chen (South China University of Technology, School of Design), Xuemiao Xu (South China University of Technology, School of Computer Science and Engineering; Guangdong Engineering Center for Large Model and GenAI Technology; State Key Laboratory of Subtropical Building and Urban Science; Ministry of Education Key Laboratory of Big Data and Intelligent Robot), Haoxin Yang (South China University of Technology, School of Computer Science and Engineering), Cheng Xu (Singapore Management University, School of Computing and Information Systems), Yihong Lin (South China University of Technology, School of Computer Science and Engineering), Shengfeng He (Singapore Management University, School of Computing and Information Systems) 💡 毒舌点评 这篇论文在解决音频驱动对话头像运动的“信号纠缠”问题上提出了一个相当清晰且有效的范式,通过分离预训练运动先验和交互调制,避免了从头学习端到端模型的复杂性和数据依赖,其核心思想和模块设计(如跨注意力交互预测)具有启发性。然而,其主要评估仅限于单一数据集(DualTalk),且模型对预训练独白模型质量的依赖程度未被充分讨论,这使得其声称的“模型无关性”和“可扩展性”缺乏更广泛的实证支撑。 ...